fs/proc/task_mmu.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/pagewalk.h>
   3 #include <linux/mm_inline.h>
   4 #include <linux/hugetlb.h>
   5 #include <linux/huge_mm.h>
   6 #include <linux/mount.h>
   7 #include <linux/ksm.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 struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv,
 128                                                 loff_t *ppos)
 129 {
 130         struct vm_area_struct *vma = vma_next(&priv->iter);
 131
 132         if (vma) {
 133                 *ppos = vma->vm_start;
 134         } else {
 135                 *ppos = -2UL;
 136                 vma = get_gate_vma(priv->mm);
 137         }
 138
 139         return vma;
 140 }
 141
 142 static void *m_start(struct seq_file *m, loff_t *ppos)
 143 {
 144         struct proc_maps_private *priv = m->private;
 145         unsigned long last_addr = *ppos;
 146         struct mm_struct *mm;
 147
 148         /* See m_next(). Zero at the start or after lseek. */
 149         if (last_addr == -1UL)
 150                 return NULL;
 151
 152         priv->task = get_proc_task(priv->inode);
 153         if (!priv->task)
 154                 return ERR_PTR(-ESRCH);
 155
 156         mm = priv->mm;
 157         if (!mm || !mmget_not_zero(mm)) {
 158                 put_task_struct(priv->task);
 159                 priv->task = NULL;
 160                 return NULL;
 161         }
 162
 163         if (mmap_read_lock_killable(mm)) {
 164                 mmput(mm);
 165                 put_task_struct(priv->task);
 166                 priv->task = NULL;
 167                 return ERR_PTR(-EINTR);
 168         }
 169
 170         vma_iter_init(&priv->iter, mm, last_addr);
 171         hold_task_mempolicy(priv);
 172         if (last_addr == -2UL)
 173                 return get_gate_vma(mm);
 174
 175         return proc_get_vma(priv, ppos);
 176 }
 177
 178 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
 179 {
 180         if (*ppos == -2UL) {
 181                 *ppos = -1UL;
 182                 return NULL;
 183         }
 184         return proc_get_vma(m->private, ppos);
 185 }
 186
 187 static void m_stop(struct seq_file *m, void *v)
 188 {
 189         struct proc_maps_private *priv = m->private;
 190         struct mm_struct *mm = priv->mm;
 191
 192         if (!priv->task)
 193                 return;
 194
 195         release_task_mempolicy(priv);
 196         mmap_read_unlock(mm);
 197         mmput(mm);
 198         put_task_struct(priv->task);
 199         priv->task = NULL;
 200 }
 201
 202 static int proc_maps_open(struct inode *inode, struct file *file,
 203                         const struct seq_operations *ops, int psize)
 204 {
 205         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 206
 207         if (!priv)
 208                 return -ENOMEM;
 209
 210         priv->inode = inode;
 211         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 212         if (IS_ERR(priv->mm)) {
 213                 int err = PTR_ERR(priv->mm);
 214
 215                 seq_release_private(inode, file);
 216                 return err;
 217         }
 218
 219         return 0;
 220 }
 221
 222 static int proc_map_release(struct inode *inode, struct file *file)
 223 {
 224         struct seq_file *seq = file->private_data;
 225         struct proc_maps_private *priv = seq->private;
 226
 227         if (priv->mm)
 228                 mmdrop(priv->mm);
 229
 230         return seq_release_private(inode, file);
 231 }
 232
 233 static int do_maps_open(struct inode *inode, struct file *file,
 234                         const struct seq_operations *ops)
 235 {
 236         return proc_maps_open(inode, file, ops,
 237                                 sizeof(struct proc_maps_private));
 238 }
 239
 240 static void show_vma_header_prefix(struct seq_file *m,
 241                                    unsigned long start, unsigned long end,
 242                                    vm_flags_t flags, unsigned long long pgoff,
 243                                    dev_t dev, unsigned long ino)
 244 {
 245         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 246         seq_put_hex_ll(m, NULL, start, 8);
 247         seq_put_hex_ll(m, "-", end, 8);
 248         seq_putc(m, ' ');
 249         seq_putc(m, flags & VM_READ ? 'r' : '-');
 250         seq_putc(m, flags & VM_WRITE ? 'w' : '-');
 251         seq_putc(m, flags & VM_EXEC ? 'x' : '-');
 252         seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
 253         seq_put_hex_ll(m, " ", pgoff, 8);
 254         seq_put_hex_ll(m, " ", MAJOR(dev), 2);
 255         seq_put_hex_ll(m, ":", MINOR(dev), 2);
 256         seq_put_decimal_ull(m, " ", ino);
 257         seq_putc(m, ' ');
 258 }
 259
 260 static void
 261 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 262 {
 263         struct anon_vma_name *anon_name = NULL;
 264         struct mm_struct *mm = vma->vm_mm;
 265         struct file *file = vma->vm_file;
 266         vm_flags_t flags = vma->vm_flags;
 267         unsigned long ino = 0;
 268         unsigned long long pgoff = 0;
 269         unsigned long start, end;
 270         dev_t dev = 0;
 271         const char *name = NULL;
 272
 273         if (file) {
 274                 struct inode *inode = file_inode(vma->vm_file);
 275                 dev = inode->i_sb->s_dev;
 276                 ino = inode->i_ino;
 277                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 278         }
 279
 280         start = vma->vm_start;
 281         end = vma->vm_end;
 282         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 283         if (mm)
 284                 anon_name = anon_vma_name(vma);
 285
 286         /*
 287          * Print the dentry name for named mappings, and a
 288          * special [heap] marker for the heap:
 289          */
 290         if (file) {
 291                 seq_pad(m, ' ');
 292                 /*
 293                  * If user named this anon shared memory via
 294                  * prctl(PR_SET_VMA ..., use the provided name.
 295                  */
 296                 if (anon_name)
 297                         seq_printf(m, "[anon_shmem:%s]", anon_name->name);
 298                 else
 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_is_initial_heap(vma)) {
 317                         name = "[heap]";
 318                         goto done;
 319                 }
 320
 321                 if (vma_is_initial_stack(vma)) {
 322                         name = "[stack]";
 323                         goto done;
 324                 }
 325
 326                 if (anon_name) {
 327                         seq_pad(m, ' ');
 328                         seq_printf(m, "[anon:%s]", anon_name->name);
 329                 }
 330         }
 331
 332 done:
 333         if (name) {
 334                 seq_pad(m, ' ');
 335                 seq_puts(m, name);
 336         }
 337         seq_putc(m, '\n');
 338 }
 339
 340 static int show_map(struct seq_file *m, void *v)
 341 {
 342         show_map_vma(m, v);
 343         return 0;
 344 }
 345
 346 static const struct seq_operations proc_pid_maps_op = {
 347         .start  = m_start,
 348         .next   = m_next,
 349         .stop   = m_stop,
 350         .show   = show_map
 351 };
 352
 353 static int pid_maps_open(struct inode *inode, struct file *file)
 354 {
 355         return do_maps_open(inode, file, &proc_pid_maps_op);
 356 }
 357
 358 const struct file_operations proc_pid_maps_operations = {
 359         .open           = pid_maps_open,
 360         .read           = seq_read,
 361         .llseek         = seq_lseek,
 362         .release        = proc_map_release,
 363 };
 364
 365 /*
 366  * Proportional Set Size(PSS): my share of RSS.
 367  *
 368  * PSS of a process is the count of pages it has in memory, where each
 369  * page is divided by the number of processes sharing it.  So if a
 370  * process has 1000 pages all to itself, and 1000 shared with one other
 371  * process, its PSS will be 1500.
 372  *
 373  * To keep (accumulated) division errors low, we adopt a 64bit
 374  * fixed-point pss counter to minimize division errors. So (pss >>
 375  * PSS_SHIFT) would be the real byte count.
 376  *
 377  * A shift of 12 before division means (assuming 4K page size):
 378  *      - 1M 3-user-pages add up to 8KB errors;
 379  *      - supports mapcount up to 2^24, or 16M;
 380  *      - supports PSS up to 2^52 bytes, or 4PB.
 381  */
 382 #define PSS_SHIFT 12
 383
 384 #ifdef CONFIG_PROC_PAGE_MONITOR
 385 struct mem_size_stats {
 386         unsigned long resident;
 387         unsigned long shared_clean;
 388         unsigned long shared_dirty;
 389         unsigned long private_clean;
 390         unsigned long private_dirty;
 391         unsigned long referenced;
 392         unsigned long anonymous;
 393         unsigned long lazyfree;
 394         unsigned long anonymous_thp;
 395         unsigned long shmem_thp;
 396         unsigned long file_thp;
 397         unsigned long swap;
 398         unsigned long shared_hugetlb;
 399         unsigned long private_hugetlb;
 400         unsigned long ksm;
 401         u64 pss;
 402         u64 pss_anon;
 403         u64 pss_file;
 404         u64 pss_shmem;
 405         u64 pss_dirty;
 406         u64 pss_locked;
 407         u64 swap_pss;
 408 };
 409
 410 static void smaps_page_accumulate(struct mem_size_stats *mss,
 411                 struct page *page, unsigned long size, unsigned long pss,
 412                 bool dirty, bool locked, bool private)
 413 {
 414         mss->pss += pss;
 415
 416         if (PageAnon(page))
 417                 mss->pss_anon += pss;
 418         else if (PageSwapBacked(page))
 419                 mss->pss_shmem += pss;
 420         else
 421                 mss->pss_file += pss;
 422
 423         if (locked)
 424                 mss->pss_locked += pss;
 425
 426         if (dirty || PageDirty(page)) {
 427                 mss->pss_dirty += pss;
 428                 if (private)
 429                         mss->private_dirty += size;
 430                 else
 431                         mss->shared_dirty += size;
 432         } else {
 433                 if (private)
 434                         mss->private_clean += size;
 435                 else
 436                         mss->shared_clean += size;
 437         }
 438 }
 439
 440 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 441                 bool compound, bool young, bool dirty, bool locked,
 442                 bool migration)
 443 {
 444         int i, nr = compound ? compound_nr(page) : 1;
 445         unsigned long size = nr * PAGE_SIZE;
 446
 447         /*
 448          * First accumulate quantities that depend only on |size| and the type
 449          * of the compound page.
 450          */
 451         if (PageAnon(page)) {
 452                 mss->anonymous += size;
 453                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 454                         mss->lazyfree += size;
 455         }
 456
 457         if (PageKsm(page))
 458                 mss->ksm += size;
 459
 460         mss->resident += size;
 461         /* Accumulate the size in pages that have been accessed. */
 462         if (young || page_is_young(page) || PageReferenced(page))
 463                 mss->referenced += size;
 464
 465         /*
 466          * Then accumulate quantities that may depend on sharing, or that may
 467          * differ page-by-page.
 468          *
 469          * page_count(page) == 1 guarantees the page is mapped exactly once.
 470          * If any subpage of the compound page mapped with PTE it would elevate
 471          * page_count().
 472          *
 473          * The page_mapcount() is called to get a snapshot of the mapcount.
 474          * Without holding the page lock this snapshot can be slightly wrong as
 475          * we cannot always read the mapcount atomically.  It is not safe to
 476          * call page_mapcount() even with PTL held if the page is not mapped,
 477          * especially for migration entries.  Treat regular migration entries
 478          * as mapcount == 1.
 479          */
 480         if ((page_count(page) == 1) || migration) {
 481                 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
 482                         locked, true);
 483                 return;
 484         }
 485         for (i = 0; i < nr; i++, page++) {
 486                 int mapcount = page_mapcount(page);
 487                 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
 488                 if (mapcount >= 2)
 489                         pss /= mapcount;
 490                 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
 491                                       mapcount < 2);
 492         }
 493 }
 494
 495 #ifdef CONFIG_SHMEM
 496 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 497                           __always_unused int depth, struct mm_walk *walk)
 498 {
 499         struct mem_size_stats *mss = walk->private;
 500         struct vm_area_struct *vma = walk->vma;
 501
 502         mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping,
 503                                               linear_page_index(vma, addr),
 504                                               linear_page_index(vma, end));
 505
 506         return 0;
 507 }
 508 #else
 509 #define smaps_pte_hole          NULL
 510 #endif /* CONFIG_SHMEM */
 511
 512 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
 513 {
 514 #ifdef CONFIG_SHMEM
 515         if (walk->ops->pte_hole) {
 516                 /* depth is not used */
 517                 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
 518         }
 519 #endif
 520 }
 521
 522 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 523                 struct mm_walk *walk)
 524 {
 525         struct mem_size_stats *mss = walk->private;
 526         struct vm_area_struct *vma = walk->vma;
 527         bool locked = !!(vma->vm_flags & VM_LOCKED);
 528         struct page *page = NULL;
 529         bool migration = false, young = false, dirty = false;
 530         pte_t ptent = ptep_get(pte);
 531
 532         if (pte_present(ptent)) {
 533                 page = vm_normal_page(vma, addr, ptent);
 534                 young = pte_young(ptent);
 535                 dirty = pte_dirty(ptent);
 536         } else if (is_swap_pte(ptent)) {
 537                 swp_entry_t swpent = pte_to_swp_entry(ptent);
 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                 page = vm_normal_page_pmd(vma, addr, *pmd);
 580         } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
 581                 swp_entry_t entry = pmd_to_swp_entry(*pmd);
 582
 583                 if (is_migration_entry(entry)) {
 584                         migration = true;
 585                         page = pfn_swap_entry_to_page(entry);
 586                 }
 587         }
 588         if (IS_ERR_OR_NULL(page))
 589                 return;
 590         if (PageAnon(page))
 591                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 592         else if (PageSwapBacked(page))
 593                 mss->shmem_thp += HPAGE_PMD_SIZE;
 594         else if (is_zone_device_page(page))
 595                 /* pass */;
 596         else
 597                 mss->file_thp += HPAGE_PMD_SIZE;
 598
 599         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
 600                       locked, migration);
 601 }
 602 #else
 603 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 604                 struct mm_walk *walk)
 605 {
 606 }
 607 #endif
 608
 609 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 610                            struct mm_walk *walk)
 611 {
 612         struct vm_area_struct *vma = walk->vma;
 613         pte_t *pte;
 614         spinlock_t *ptl;
 615
 616         ptl = pmd_trans_huge_lock(pmd, vma);
 617         if (ptl) {
 618                 smaps_pmd_entry(pmd, addr, walk);
 619                 spin_unlock(ptl);
 620                 goto out;
 621         }
 622
 623         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 624         if (!pte) {
 625                 walk->action = ACTION_AGAIN;
 626                 return 0;
 627         }
 628         for (; addr != end; pte++, addr += PAGE_SIZE)
 629                 smaps_pte_entry(pte, addr, walk);
 630         pte_unmap_unlock(pte - 1, ptl);
 631 out:
 632         cond_resched();
 633         return 0;
 634 }
 635
 636 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 637 {
 638         /*
 639          * Don't forget to update Documentation/ on changes.
 640          */
 641         static const char mnemonics[BITS_PER_LONG][2] = {
 642                 /*
 643                  * In case if we meet a flag we don't know about.
 644                  */
 645                 [0 ... (BITS_PER_LONG-1)] = "??",
 646
 647                 [ilog2(VM_READ)]        = "rd",
 648                 [ilog2(VM_WRITE)]       = "wr",
 649                 [ilog2(VM_EXEC)]        = "ex",
 650                 [ilog2(VM_SHARED)]      = "sh",
 651                 [ilog2(VM_MAYREAD)]     = "mr",
 652                 [ilog2(VM_MAYWRITE)]    = "mw",
 653                 [ilog2(VM_MAYEXEC)]     = "me",
 654                 [ilog2(VM_MAYSHARE)]    = "ms",
 655                 [ilog2(VM_GROWSDOWN)]   = "gd",
 656                 [ilog2(VM_PFNMAP)]      = "pf",
 657                 [ilog2(VM_LOCKED)]      = "lo",
 658                 [ilog2(VM_IO)]          = "io",
 659                 [ilog2(VM_SEQ_READ)]    = "sr",
 660                 [ilog2(VM_RAND_READ)]   = "rr",
 661                 [ilog2(VM_DONTCOPY)]    = "dc",
 662                 [ilog2(VM_DONTEXPAND)]  = "de",
 663                 [ilog2(VM_LOCKONFAULT)] = "lf",
 664                 [ilog2(VM_ACCOUNT)]     = "ac",
 665                 [ilog2(VM_NORESERVE)]   = "nr",
 666                 [ilog2(VM_HUGETLB)]     = "ht",
 667                 [ilog2(VM_SYNC)]        = "sf",
 668                 [ilog2(VM_ARCH_1)]      = "ar",
 669                 [ilog2(VM_WIPEONFORK)]  = "wf",
 670                 [ilog2(VM_DONTDUMP)]    = "dd",
 671 #ifdef CONFIG_ARM64_BTI
 672                 [ilog2(VM_ARM64_BTI)]   = "bt",
 673 #endif
 674 #ifdef CONFIG_MEM_SOFT_DIRTY
 675                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 676 #endif
 677                 [ilog2(VM_MIXEDMAP)]    = "mm",
 678                 [ilog2(VM_HUGEPAGE)]    = "hg",
 679                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 680                 [ilog2(VM_MERGEABLE)]   = "mg",
 681                 [ilog2(VM_UFFD_MISSING)]= "um",
 682                 [ilog2(VM_UFFD_WP)]     = "uw",
 683 #ifdef CONFIG_ARM64_MTE
 684                 [ilog2(VM_MTE)]         = "mt",
 685                 [ilog2(VM_MTE_ALLOWED)] = "",
 686 #endif
 687 #ifdef CONFIG_ARCH_HAS_PKEYS
 688                 /* These come out via ProtectionKey: */
 689                 [ilog2(VM_PKEY_BIT0)]   = "",
 690                 [ilog2(VM_PKEY_BIT1)]   = "",
 691                 [ilog2(VM_PKEY_BIT2)]   = "",
 692                 [ilog2(VM_PKEY_BIT3)]   = "",
 693 #if VM_PKEY_BIT4
 694                 [ilog2(VM_PKEY_BIT4)]   = "",
 695 #endif
 696 #endif /* CONFIG_ARCH_HAS_PKEYS */
 697 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
 698                 [ilog2(VM_UFFD_MINOR)]  = "ui",
 699 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
 700 #ifdef CONFIG_X86_USER_SHADOW_STACK
 701                 [ilog2(VM_SHADOW_STACK)] = "ss",
 702 #endif
 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         pte_t ptent = ptep_get(pte);
 728
 729         if (pte_present(ptent)) {
 730                 page = vm_normal_page(vma, addr, ptent);
 731         } else if (is_swap_pte(ptent)) {
 732                 swp_entry_t swpent = pte_to_swp_entry(ptent);
 733
 734                 if (is_pfn_swap_entry(swpent))
 735                         page = pfn_swap_entry_to_page(swpent);
 736         }
 737         if (page) {
 738                 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
 739                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 740                 else
 741                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 742         }
 743         return 0;
 744 }
 745 #else
 746 #define smaps_hugetlb_range     NULL
 747 #endif /* HUGETLB_PAGE */
 748
 749 static const struct mm_walk_ops smaps_walk_ops = {
 750         .pmd_entry              = smaps_pte_range,
 751         .hugetlb_entry          = smaps_hugetlb_range,
 752         .walk_lock              = PGWALK_RDLOCK,
 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         .walk_lock              = PGWALK_RDLOCK,
 760 };
 761
 762 /*
 763  * Gather mem stats from @vma with the indicated beginning
 764  * address @start, and keep them in @mss.
 765  *
 766  * Use vm_start of @vma as the beginning address if @start is 0.
 767  */
 768 static void smap_gather_stats(struct vm_area_struct *vma,
 769                 struct mem_size_stats *mss, unsigned long start)
 770 {
 771         const struct mm_walk_ops *ops = &smaps_walk_ops;
 772
 773         /* Invalid start */
 774         if (start >= vma->vm_end)
 775                 return;
 776
 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
 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\nKSM:            ", mss->ksm);
 834         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 835         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 836         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 837         SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
 838         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 839         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 840                                   mss->private_hugetlb >> 10, 7);
 841         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 842         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 843                                         mss->swap_pss >> PSS_SHIFT);
 844         SEQ_PUT_DEC(" kB\nLocked:         ",
 845                                         mss->pss_locked >> PSS_SHIFT);
 846         seq_puts(m, " kB\n");
 847 }
 848
 849 static int show_smap(struct seq_file *m, void *v)
 850 {
 851         struct vm_area_struct *vma = v;
 852         struct mem_size_stats mss = {};
 853
 854         smap_gather_stats(vma, &mss, 0);
 855
 856         show_map_vma(m, vma);
 857
 858         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 859         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 860         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 861         seq_puts(m, " kB\n");
 862
 863         __show_smap(m, &mss, false);
 864
 865         seq_printf(m, "THPeligible:    %8u\n",
 866                    hugepage_vma_check(vma, vma->vm_flags, true, false, true));
 867
 868         if (arch_pkeys_enabled())
 869                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 870         show_smap_vma_flags(m, vma);
 871
 872         return 0;
 873 }
 874
 875 static int show_smaps_rollup(struct seq_file *m, void *v)
 876 {
 877         struct proc_maps_private *priv = m->private;
 878         struct mem_size_stats mss = {};
 879         struct mm_struct *mm = priv->mm;
 880         struct vm_area_struct *vma;
 881         unsigned long vma_start = 0, last_vma_end = 0;
 882         int ret = 0;
 883         VMA_ITERATOR(vmi, mm, 0);
 884
 885         priv->task = get_proc_task(priv->inode);
 886         if (!priv->task)
 887                 return -ESRCH;
 888
 889         if (!mm || !mmget_not_zero(mm)) {
 890                 ret = -ESRCH;
 891                 goto out_put_task;
 892         }
 893
 894         ret = mmap_read_lock_killable(mm);
 895         if (ret)
 896                 goto out_put_mm;
 897
 898         hold_task_mempolicy(priv);
 899         vma = vma_next(&vmi);
 900
 901         if (unlikely(!vma))
 902                 goto empty_set;
 903
 904         vma_start = vma->vm_start;
 905         do {
 906                 smap_gather_stats(vma, &mss, 0);
 907                 last_vma_end = vma->vm_end;
 908
 909                 /*
 910                  * Release mmap_lock temporarily if someone wants to
 911                  * access it for write request.
 912                  */
 913                 if (mmap_lock_is_contended(mm)) {
 914                         vma_iter_invalidate(&vmi);
 915                         mmap_read_unlock(mm);
 916                         ret = mmap_read_lock_killable(mm);
 917                         if (ret) {
 918                                 release_task_mempolicy(priv);
 919                                 goto out_put_mm;
 920                         }
 921
 922                         /*
 923                          * After dropping the lock, there are four cases to
 924                          * consider. See the following example for explanation.
 925                          *
 926                          *   +------+------+-----------+
 927                          *   | VMA1 | VMA2 | VMA3      |
 928                          *   +------+------+-----------+
 929                          *   |      |      |           |
 930                          *  4k     8k     16k         400k
 931                          *
 932                          * Suppose we drop the lock after reading VMA2 due to
 933                          * contention, then we get:
 934                          *
 935                          *      last_vma_end = 16k
 936                          *
 937                          * 1) VMA2 is freed, but VMA3 exists:
 938                          *
 939                          *    vma_next(vmi) will return VMA3.
 940                          *    In this case, just continue from VMA3.
 941                          *
 942                          * 2) VMA2 still exists:
 943                          *
 944                          *    vma_next(vmi) will return VMA3.
 945                          *    In this case, just continue from VMA3.
 946                          *
 947                          * 3) No more VMAs can be found:
 948                          *
 949                          *    vma_next(vmi) will return NULL.
 950                          *    No more things to do, just break.
 951                          *
 952                          * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
 953                          *
 954                          *    vma_next(vmi) will return VMA' whose range
 955                          *    contains last_vma_end.
 956                          *    Iterate VMA' from last_vma_end.
 957                          */
 958                         vma = vma_next(&vmi);
 959                         /* Case 3 above */
 960                         if (!vma)
 961                                 break;
 962
 963                         /* Case 1 and 2 above */
 964                         if (vma->vm_start >= last_vma_end)
 965                                 continue;
 966
 967                         /* Case 4 above */
 968                         if (vma->vm_end > last_vma_end)
 969                                 smap_gather_stats(vma, &mss, last_vma_end);
 970                 }
 971         } for_each_vma(vmi, vma);
 972
 973 empty_set:
 974         show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
 975         seq_pad(m, ' ');
 976         seq_puts(m, "[rollup]\n");
 977
 978         __show_smap(m, &mss, true);
 979
 980         release_task_mempolicy(priv);
 981         mmap_read_unlock(mm);
 982
 983 out_put_mm:
 984         mmput(mm);
 985 out_put_task:
 986         put_task_struct(priv->task);
 987         priv->task = NULL;
 988
 989         return ret;
 990 }
 991 #undef SEQ_PUT_DEC
 992
 993 static const struct seq_operations proc_pid_smaps_op = {
 994         .start  = m_start,
 995         .next   = m_next,
 996         .stop   = m_stop,
 997         .show   = show_smap
 998 };
 999
1000 static int pid_smaps_open(struct inode *inode, struct file *file)
1001 {
1002         return do_maps_open(inode, file, &proc_pid_smaps_op);
1003 }
1004
1005 static int smaps_rollup_open(struct inode *inode, struct file *file)
1006 {
1007         int ret;
1008         struct proc_maps_private *priv;
1009
1010         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1011         if (!priv)
1012                 return -ENOMEM;
1013
1014         ret = single_open(file, show_smaps_rollup, priv);
1015         if (ret)
1016                 goto out_free;
1017
1018         priv->inode = inode;
1019         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1020         if (IS_ERR(priv->mm)) {
1021                 ret = PTR_ERR(priv->mm);
1022
1023                 single_release(inode, file);
1024                 goto out_free;
1025         }
1026
1027         return 0;
1028
1029 out_free:
1030         kfree(priv);
1031         return ret;
1032 }
1033
1034 static int smaps_rollup_release(struct inode *inode, struct file *file)
1035 {
1036         struct seq_file *seq = file->private_data;
1037         struct proc_maps_private *priv = seq->private;
1038
1039         if (priv->mm)
1040                 mmdrop(priv->mm);
1041
1042         kfree(priv);
1043         return single_release(inode, file);
1044 }
1045
1046 const struct file_operations proc_pid_smaps_operations = {
1047         .open           = pid_smaps_open,
1048         .read           = seq_read,
1049         .llseek         = seq_lseek,
1050         .release        = proc_map_release,
1051 };
1052
1053 const struct file_operations proc_pid_smaps_rollup_operations = {
1054         .open           = smaps_rollup_open,
1055         .read           = seq_read,
1056         .llseek         = seq_lseek,
1057         .release        = smaps_rollup_release,
1058 };
1059
1060 enum clear_refs_types {
1061         CLEAR_REFS_ALL = 1,
1062         CLEAR_REFS_ANON,
1063         CLEAR_REFS_MAPPED,
1064         CLEAR_REFS_SOFT_DIRTY,
1065         CLEAR_REFS_MM_HIWATER_RSS,
1066         CLEAR_REFS_LAST,
1067 };
1068
1069 struct clear_refs_private {
1070         enum clear_refs_types type;
1071 };
1072
1073 #ifdef CONFIG_MEM_SOFT_DIRTY
1074
1075 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1076 {
1077         struct page *page;
1078
1079         if (!pte_write(pte))
1080                 return false;
1081         if (!is_cow_mapping(vma->vm_flags))
1082                 return false;
1083         if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1084                 return false;
1085         page = vm_normal_page(vma, addr, pte);
1086         if (!page)
1087                 return false;
1088         return page_maybe_dma_pinned(page);
1089 }
1090
1091 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1092                 unsigned long addr, pte_t *pte)
1093 {
1094         /*
1095          * The soft-dirty tracker uses #PF-s to catch writes
1096          * to pages, so write-protect the pte as well. See the
1097          * Documentation/admin-guide/mm/soft-dirty.rst for full description
1098          * of how soft-dirty works.
1099          */
1100         pte_t ptent = ptep_get(pte);
1101
1102         if (pte_present(ptent)) {
1103                 pte_t old_pte;
1104
1105                 if (pte_is_pinned(vma, addr, ptent))
1106                         return;
1107                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1108                 ptent = pte_wrprotect(old_pte);
1109                 ptent = pte_clear_soft_dirty(ptent);
1110                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1111         } else if (is_swap_pte(ptent)) {
1112                 ptent = pte_swp_clear_soft_dirty(ptent);
1113                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1114         }
1115 }
1116 #else
1117 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1118                 unsigned long addr, pte_t *pte)
1119 {
1120 }
1121 #endif
1122
1123 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1124 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1125                 unsigned long addr, pmd_t *pmdp)
1126 {
1127         pmd_t old, pmd = *pmdp;
1128
1129         if (pmd_present(pmd)) {
1130                 /* See comment in change_huge_pmd() */
1131                 old = pmdp_invalidate(vma, addr, pmdp);
1132                 if (pmd_dirty(old))
1133                         pmd = pmd_mkdirty(pmd);
1134                 if (pmd_young(old))
1135                         pmd = pmd_mkyoung(pmd);
1136
1137                 pmd = pmd_wrprotect(pmd);
1138                 pmd = pmd_clear_soft_dirty(pmd);
1139
1140                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1141         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1142                 pmd = pmd_swp_clear_soft_dirty(pmd);
1143                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1144         }
1145 }
1146 #else
1147 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1148                 unsigned long addr, pmd_t *pmdp)
1149 {
1150 }
1151 #endif
1152
1153 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1154                                 unsigned long end, struct mm_walk *walk)
1155 {
1156         struct clear_refs_private *cp = walk->private;
1157         struct vm_area_struct *vma = walk->vma;
1158         pte_t *pte, ptent;
1159         spinlock_t *ptl;
1160         struct page *page;
1161
1162         ptl = pmd_trans_huge_lock(pmd, vma);
1163         if (ptl) {
1164                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1165                         clear_soft_dirty_pmd(vma, addr, pmd);
1166                         goto out;
1167                 }
1168
1169                 if (!pmd_present(*pmd))
1170                         goto out;
1171
1172                 page = pmd_page(*pmd);
1173
1174                 /* Clear accessed and referenced bits. */
1175                 pmdp_test_and_clear_young(vma, addr, pmd);
1176                 test_and_clear_page_young(page);
1177                 ClearPageReferenced(page);
1178 out:
1179                 spin_unlock(ptl);
1180                 return 0;
1181         }
1182
1183         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1184         if (!pte) {
1185                 walk->action = ACTION_AGAIN;
1186                 return 0;
1187         }
1188         for (; addr != end; pte++, addr += PAGE_SIZE) {
1189                 ptent = ptep_get(pte);
1190
1191                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1192                         clear_soft_dirty(vma, addr, pte);
1193                         continue;
1194                 }
1195
1196                 if (!pte_present(ptent))
1197                         continue;
1198
1199                 page = vm_normal_page(vma, addr, ptent);
1200                 if (!page)
1201                         continue;
1202
1203                 /* Clear accessed and referenced bits. */
1204                 ptep_test_and_clear_young(vma, addr, pte);
1205                 test_and_clear_page_young(page);
1206                 ClearPageReferenced(page);
1207         }
1208         pte_unmap_unlock(pte - 1, ptl);
1209         cond_resched();
1210         return 0;
1211 }
1212
1213 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1214                                 struct mm_walk *walk)
1215 {
1216         struct clear_refs_private *cp = walk->private;
1217         struct vm_area_struct *vma = walk->vma;
1218
1219         if (vma->vm_flags & VM_PFNMAP)
1220                 return 1;
1221
1222         /*
1223          * Writing 1 to /proc/pid/clear_refs affects all pages.
1224          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1225          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1226          * Writing 4 to /proc/pid/clear_refs affects all pages.
1227          */
1228         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1229                 return 1;
1230         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1231                 return 1;
1232         return 0;
1233 }
1234
1235 static const struct mm_walk_ops clear_refs_walk_ops = {
1236         .pmd_entry              = clear_refs_pte_range,
1237         .test_walk              = clear_refs_test_walk,
1238         .walk_lock              = PGWALK_WRLOCK,
1239 };
1240
1241 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1242                                 size_t count, loff_t *ppos)
1243 {
1244         struct task_struct *task;
1245         char buffer[PROC_NUMBUF] = {};
1246         struct mm_struct *mm;
1247         struct vm_area_struct *vma;
1248         enum clear_refs_types type;
1249         int itype;
1250         int rv;
1251
1252         if (count > sizeof(buffer) - 1)
1253                 count = sizeof(buffer) - 1;
1254         if (copy_from_user(buffer, buf, count))
1255                 return -EFAULT;
1256         rv = kstrtoint(strstrip(buffer), 10, &itype);
1257         if (rv < 0)
1258                 return rv;
1259         type = (enum clear_refs_types)itype;
1260         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1261                 return -EINVAL;
1262
1263         task = get_proc_task(file_inode(file));
1264         if (!task)
1265                 return -ESRCH;
1266         mm = get_task_mm(task);
1267         if (mm) {
1268                 VMA_ITERATOR(vmi, mm, 0);
1269                 struct mmu_notifier_range range;
1270                 struct clear_refs_private cp = {
1271                         .type = type,
1272                 };
1273
1274                 if (mmap_write_lock_killable(mm)) {
1275                         count = -EINTR;
1276                         goto out_mm;
1277                 }
1278                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1279                         /*
1280                          * Writing 5 to /proc/pid/clear_refs resets the peak
1281                          * resident set size to this mm's current rss value.
1282                          */
1283                         reset_mm_hiwater_rss(mm);
1284                         goto out_unlock;
1285                 }
1286
1287                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1288                         for_each_vma(vmi, vma) {
1289                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1290                                         continue;
1291                                 vm_flags_clear(vma, VM_SOFTDIRTY);
1292                                 vma_set_page_prot(vma);
1293                         }
1294
1295                         inc_tlb_flush_pending(mm);
1296                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1297                                                 0, mm, 0, -1UL);
1298                         mmu_notifier_invalidate_range_start(&range);
1299                 }
1300                 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1301                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1302                         mmu_notifier_invalidate_range_end(&range);
1303                         flush_tlb_mm(mm);
1304                         dec_tlb_flush_pending(mm);
1305                 }
1306 out_unlock:
1307                 mmap_write_unlock(mm);
1308 out_mm:
1309                 mmput(mm);
1310         }
1311         put_task_struct(task);
1312
1313         return count;
1314 }
1315
1316 const struct file_operations proc_clear_refs_operations = {
1317         .write          = clear_refs_write,
1318         .llseek         = noop_llseek,
1319 };
1320
1321 typedef struct {
1322         u64 pme;
1323 } pagemap_entry_t;
1324
1325 struct pagemapread {
1326         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1327         pagemap_entry_t *buffer;
1328         bool show_pfn;
1329 };
1330
1331 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1332 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1333
1334 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1335 #define PM_PFRAME_BITS          55
1336 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1337 #define PM_SOFT_DIRTY           BIT_ULL(55)
1338 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1339 #define PM_UFFD_WP              BIT_ULL(57)
1340 #define PM_FILE                 BIT_ULL(61)
1341 #define PM_SWAP                 BIT_ULL(62)
1342 #define PM_PRESENT              BIT_ULL(63)
1343
1344 #define PM_END_OF_BUFFER    1
1345
1346 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1347 {
1348         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1349 }
1350
1351 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1352                           struct pagemapread *pm)
1353 {
1354         pm->buffer[pm->pos++] = *pme;
1355         if (pm->pos >= pm->len)
1356                 return PM_END_OF_BUFFER;
1357         return 0;
1358 }
1359
1360 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1361                             __always_unused int depth, struct mm_walk *walk)
1362 {
1363         struct pagemapread *pm = walk->private;
1364         unsigned long addr = start;
1365         int err = 0;
1366
1367         while (addr < end) {
1368                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1369                 pagemap_entry_t pme = make_pme(0, 0);
1370                 /* End of address space hole, which we mark as non-present. */
1371                 unsigned long hole_end;
1372
1373                 if (vma)
1374                         hole_end = min(end, vma->vm_start);
1375                 else
1376                         hole_end = end;
1377
1378                 for (; addr < hole_end; addr += PAGE_SIZE) {
1379                         err = add_to_pagemap(addr, &pme, pm);
1380                         if (err)
1381                                 goto out;
1382                 }
1383
1384                 if (!vma)
1385                         break;
1386
1387                 /* Addresses in the VMA. */
1388                 if (vma->vm_flags & VM_SOFTDIRTY)
1389                         pme = make_pme(0, PM_SOFT_DIRTY);
1390                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1391                         err = add_to_pagemap(addr, &pme, pm);
1392                         if (err)
1393                                 goto out;
1394                 }
1395         }
1396 out:
1397         return err;
1398 }
1399
1400 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1401                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1402 {
1403         u64 frame = 0, flags = 0;
1404         struct page *page = NULL;
1405         bool migration = false;
1406
1407         if (pte_present(pte)) {
1408                 if (pm->show_pfn)
1409                         frame = pte_pfn(pte);
1410                 flags |= PM_PRESENT;
1411                 page = vm_normal_page(vma, addr, pte);
1412                 if (pte_soft_dirty(pte))
1413                         flags |= PM_SOFT_DIRTY;
1414                 if (pte_uffd_wp(pte))
1415                         flags |= PM_UFFD_WP;
1416         } else if (is_swap_pte(pte)) {
1417                 swp_entry_t entry;
1418                 if (pte_swp_soft_dirty(pte))
1419                         flags |= PM_SOFT_DIRTY;
1420                 if (pte_swp_uffd_wp(pte))
1421                         flags |= PM_UFFD_WP;
1422                 entry = pte_to_swp_entry(pte);
1423                 if (pm->show_pfn) {
1424                         pgoff_t offset;
1425                         /*
1426                          * For PFN swap offsets, keeping the offset field
1427                          * to be PFN only to be compatible with old smaps.
1428                          */
1429                         if (is_pfn_swap_entry(entry))
1430                                 offset = swp_offset_pfn(entry);
1431                         else
1432                                 offset = swp_offset(entry);
1433                         frame = swp_type(entry) |
1434                             (offset << MAX_SWAPFILES_SHIFT);
1435                 }
1436                 flags |= PM_SWAP;
1437                 migration = is_migration_entry(entry);
1438                 if (is_pfn_swap_entry(entry))
1439                         page = pfn_swap_entry_to_page(entry);
1440                 if (pte_marker_entry_uffd_wp(entry))
1441                         flags |= PM_UFFD_WP;
1442         }
1443
1444         if (page && !PageAnon(page))
1445                 flags |= PM_FILE;
1446         if (page && !migration && page_mapcount(page) == 1)
1447                 flags |= PM_MMAP_EXCLUSIVE;
1448         if (vma->vm_flags & VM_SOFTDIRTY)
1449                 flags |= PM_SOFT_DIRTY;
1450
1451         return make_pme(frame, flags);
1452 }
1453
1454 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1455                              struct mm_walk *walk)
1456 {
1457         struct vm_area_struct *vma = walk->vma;
1458         struct pagemapread *pm = walk->private;
1459         spinlock_t *ptl;
1460         pte_t *pte, *orig_pte;
1461         int err = 0;
1462 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1463         bool migration = false;
1464
1465         ptl = pmd_trans_huge_lock(pmdp, vma);
1466         if (ptl) {
1467                 u64 flags = 0, frame = 0;
1468                 pmd_t pmd = *pmdp;
1469                 struct page *page = NULL;
1470
1471                 if (vma->vm_flags & VM_SOFTDIRTY)
1472                         flags |= PM_SOFT_DIRTY;
1473
1474                 if (pmd_present(pmd)) {
1475                         page = pmd_page(pmd);
1476
1477                         flags |= PM_PRESENT;
1478                         if (pmd_soft_dirty(pmd))
1479                                 flags |= PM_SOFT_DIRTY;
1480                         if (pmd_uffd_wp(pmd))
1481                                 flags |= PM_UFFD_WP;
1482                         if (pm->show_pfn)
1483                                 frame = pmd_pfn(pmd) +
1484                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1485                 }
1486 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1487                 else if (is_swap_pmd(pmd)) {
1488                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1489                         unsigned long offset;
1490
1491                         if (pm->show_pfn) {
1492                                 if (is_pfn_swap_entry(entry))
1493                                         offset = swp_offset_pfn(entry);
1494                                 else
1495                                         offset = swp_offset(entry);
1496                                 offset = offset +
1497                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1498                                 frame = swp_type(entry) |
1499                                         (offset << MAX_SWAPFILES_SHIFT);
1500                         }
1501                         flags |= PM_SWAP;
1502                         if (pmd_swp_soft_dirty(pmd))
1503                                 flags |= PM_SOFT_DIRTY;
1504                         if (pmd_swp_uffd_wp(pmd))
1505                                 flags |= PM_UFFD_WP;
1506                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1507                         migration = is_migration_entry(entry);
1508                         page = pfn_swap_entry_to_page(entry);
1509                 }
1510 #endif
1511
1512                 if (page && !migration && page_mapcount(page) == 1)
1513                         flags |= PM_MMAP_EXCLUSIVE;
1514
1515                 for (; addr != end; addr += PAGE_SIZE) {
1516                         pagemap_entry_t pme = make_pme(frame, flags);
1517
1518                         err = add_to_pagemap(addr, &pme, pm);
1519                         if (err)
1520                                 break;
1521                         if (pm->show_pfn) {
1522                                 if (flags & PM_PRESENT)
1523                                         frame++;
1524                                 else if (flags & PM_SWAP)
1525                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1526                         }
1527                 }
1528                 spin_unlock(ptl);
1529                 return err;
1530         }
1531 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1532
1533         /*
1534          * We can assume that @vma always points to a valid one and @end never
1535          * goes beyond vma->vm_end.
1536          */
1537         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1538         if (!pte) {
1539                 walk->action = ACTION_AGAIN;
1540                 return err;
1541         }
1542         for (; addr < end; pte++, addr += PAGE_SIZE) {
1543                 pagemap_entry_t pme;
1544
1545                 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte));
1546                 err = add_to_pagemap(addr, &pme, pm);
1547                 if (err)
1548                         break;
1549         }
1550         pte_unmap_unlock(orig_pte, ptl);
1551
1552         cond_resched();
1553
1554         return err;
1555 }
1556
1557 #ifdef CONFIG_HUGETLB_PAGE
1558 /* This function walks within one hugetlb entry in the single call */
1559 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1560                                  unsigned long addr, unsigned long end,
1561                                  struct mm_walk *walk)
1562 {
1563         struct pagemapread *pm = walk->private;
1564         struct vm_area_struct *vma = walk->vma;
1565         u64 flags = 0, frame = 0;
1566         int err = 0;
1567         pte_t pte;
1568
1569         if (vma->vm_flags & VM_SOFTDIRTY)
1570                 flags |= PM_SOFT_DIRTY;
1571
1572         pte = huge_ptep_get(ptep);
1573         if (pte_present(pte)) {
1574                 struct page *page = pte_page(pte);
1575
1576                 if (!PageAnon(page))
1577                         flags |= PM_FILE;
1578
1579                 if (page_mapcount(page) == 1)
1580                         flags |= PM_MMAP_EXCLUSIVE;
1581
1582                 if (huge_pte_uffd_wp(pte))
1583                         flags |= PM_UFFD_WP;
1584
1585                 flags |= PM_PRESENT;
1586                 if (pm->show_pfn)
1587                         frame = pte_pfn(pte) +
1588                                 ((addr & ~hmask) >> PAGE_SHIFT);
1589         } else if (pte_swp_uffd_wp_any(pte)) {
1590                 flags |= PM_UFFD_WP;
1591         }
1592
1593         for (; addr != end; addr += PAGE_SIZE) {
1594                 pagemap_entry_t pme = make_pme(frame, flags);
1595
1596                 err = add_to_pagemap(addr, &pme, pm);
1597                 if (err)
1598                         return err;
1599                 if (pm->show_pfn && (flags & PM_PRESENT))
1600                         frame++;
1601         }
1602
1603         cond_resched();
1604
1605         return err;
1606 }
1607 #else
1608 #define pagemap_hugetlb_range   NULL
1609 #endif /* HUGETLB_PAGE */
1610
1611 static const struct mm_walk_ops pagemap_ops = {
1612         .pmd_entry      = pagemap_pmd_range,
1613         .pte_hole       = pagemap_pte_hole,
1614         .hugetlb_entry  = pagemap_hugetlb_range,
1615         .walk_lock      = PGWALK_RDLOCK,
1616 };
1617
1618 /*
1619  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1620  *
1621  * For each page in the address space, this file contains one 64-bit entry
1622  * consisting of the following:
1623  *
1624  * Bits 0-54  page frame number (PFN) if present
1625  * Bits 0-4   swap type if swapped
1626  * Bits 5-54  swap offset if swapped
1627  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1628  * Bit  56    page exclusively mapped
1629  * Bit  57    pte is uffd-wp write-protected
1630  * Bits 58-60 zero
1631  * Bit  61    page is file-page or shared-anon
1632  * Bit  62    page swapped
1633  * Bit  63    page present
1634  *
1635  * If the page is not present but in swap, then the PFN contains an
1636  * encoding of the swap file number and the page's offset into the
1637  * swap. Unmapped pages return a null PFN. This allows determining
1638  * precisely which pages are mapped (or in swap) and comparing mapped
1639  * pages between processes.
1640  *
1641  * Efficient users of this interface will use /proc/pid/maps to
1642  * determine which areas of memory are actually mapped and llseek to
1643  * skip over unmapped regions.
1644  */
1645 static ssize_t pagemap_read(struct file *file, char __user *buf,
1646                             size_t count, loff_t *ppos)
1647 {
1648         struct mm_struct *mm = file->private_data;
1649         struct pagemapread pm;
1650         unsigned long src;
1651         unsigned long svpfn;
1652         unsigned long start_vaddr;
1653         unsigned long end_vaddr;
1654         int ret = 0, copied = 0;
1655
1656         if (!mm || !mmget_not_zero(mm))
1657                 goto out;
1658
1659         ret = -EINVAL;
1660         /* file position must be aligned */
1661         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1662                 goto out_mm;
1663
1664         ret = 0;
1665         if (!count)
1666                 goto out_mm;
1667
1668         /* do not disclose physical addresses: attack vector */
1669         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1670
1671         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1672         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1673         ret = -ENOMEM;
1674         if (!pm.buffer)
1675                 goto out_mm;
1676
1677         src = *ppos;
1678         svpfn = src / PM_ENTRY_BYTES;
1679         end_vaddr = mm->task_size;
1680
1681         /* watch out for wraparound */
1682         start_vaddr = end_vaddr;
1683         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
1684                 unsigned long end;
1685
1686                 ret = mmap_read_lock_killable(mm);
1687                 if (ret)
1688                         goto out_free;
1689                 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
1690                 mmap_read_unlock(mm);
1691
1692                 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
1693                 if (end >= start_vaddr && end < mm->task_size)
1694                         end_vaddr = end;
1695         }
1696
1697         /* Ensure the address is inside the task */
1698         if (start_vaddr > mm->task_size)
1699                 start_vaddr = end_vaddr;
1700
1701         ret = 0;
1702         while (count && (start_vaddr < end_vaddr)) {
1703                 int len;
1704                 unsigned long end;
1705
1706                 pm.pos = 0;
1707                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1708                 /* overflow ? */
1709                 if (end < start_vaddr || end > end_vaddr)
1710                         end = end_vaddr;
1711                 ret = mmap_read_lock_killable(mm);
1712                 if (ret)
1713                         goto out_free;
1714                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1715                 mmap_read_unlock(mm);
1716                 start_vaddr = end;
1717
1718                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1719                 if (copy_to_user(buf, pm.buffer, len)) {
1720                         ret = -EFAULT;
1721                         goto out_free;
1722                 }
1723                 copied += len;
1724                 buf += len;
1725                 count -= len;
1726         }
1727         *ppos += copied;
1728         if (!ret || ret == PM_END_OF_BUFFER)
1729                 ret = copied;
1730
1731 out_free:
1732         kfree(pm.buffer);
1733 out_mm:
1734         mmput(mm);
1735 out:
1736         return ret;
1737 }
1738
1739 static int pagemap_open(struct inode *inode, struct file *file)
1740 {
1741         struct mm_struct *mm;
1742
1743         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1744         if (IS_ERR(mm))
1745                 return PTR_ERR(mm);
1746         file->private_data = mm;
1747         return 0;
1748 }
1749
1750 static int pagemap_release(struct inode *inode, struct file *file)
1751 {
1752         struct mm_struct *mm = file->private_data;
1753
1754         if (mm)
1755                 mmdrop(mm);
1756         return 0;
1757 }
1758
1759 const struct file_operations proc_pagemap_operations = {
1760         .llseek         = mem_lseek, /* borrow this */
1761         .read           = pagemap_read,
1762         .open           = pagemap_open,
1763         .release        = pagemap_release,
1764 };
1765 #endif /* CONFIG_PROC_PAGE_MONITOR */
1766
1767 #ifdef CONFIG_NUMA
1768
1769 struct numa_maps {
1770         unsigned long pages;
1771         unsigned long anon;
1772         unsigned long active;
1773         unsigned long writeback;
1774         unsigned long mapcount_max;
1775         unsigned long dirty;
1776         unsigned long swapcache;
1777         unsigned long node[MAX_NUMNODES];
1778 };
1779
1780 struct numa_maps_private {
1781         struct proc_maps_private proc_maps;
1782         struct numa_maps md;
1783 };
1784
1785 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1786                         unsigned long nr_pages)
1787 {
1788         int count = page_mapcount(page);
1789
1790         md->pages += nr_pages;
1791         if (pte_dirty || PageDirty(page))
1792                 md->dirty += nr_pages;
1793
1794         if (PageSwapCache(page))
1795                 md->swapcache += nr_pages;
1796
1797         if (PageActive(page) || PageUnevictable(page))
1798                 md->active += nr_pages;
1799
1800         if (PageWriteback(page))
1801                 md->writeback += nr_pages;
1802
1803         if (PageAnon(page))
1804                 md->anon += nr_pages;
1805
1806         if (count > md->mapcount_max)
1807                 md->mapcount_max = count;
1808
1809         md->node[page_to_nid(page)] += nr_pages;
1810 }
1811
1812 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1813                 unsigned long addr)
1814 {
1815         struct page *page;
1816         int nid;
1817
1818         if (!pte_present(pte))
1819                 return NULL;
1820
1821         page = vm_normal_page(vma, addr, pte);
1822         if (!page || is_zone_device_page(page))
1823                 return NULL;
1824
1825         if (PageReserved(page))
1826                 return NULL;
1827
1828         nid = page_to_nid(page);
1829         if (!node_isset(nid, node_states[N_MEMORY]))
1830                 return NULL;
1831
1832         return page;
1833 }
1834
1835 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1836 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1837                                               struct vm_area_struct *vma,
1838                                               unsigned long addr)
1839 {
1840         struct page *page;
1841         int nid;
1842
1843         if (!pmd_present(pmd))
1844                 return NULL;
1845
1846         page = vm_normal_page_pmd(vma, addr, pmd);
1847         if (!page)
1848                 return NULL;
1849
1850         if (PageReserved(page))
1851                 return NULL;
1852
1853         nid = page_to_nid(page);
1854         if (!node_isset(nid, node_states[N_MEMORY]))
1855                 return NULL;
1856
1857         return page;
1858 }
1859 #endif
1860
1861 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1862                 unsigned long end, struct mm_walk *walk)
1863 {
1864         struct numa_maps *md = walk->private;
1865         struct vm_area_struct *vma = walk->vma;
1866         spinlock_t *ptl;
1867         pte_t *orig_pte;
1868         pte_t *pte;
1869
1870 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1871         ptl = pmd_trans_huge_lock(pmd, vma);
1872         if (ptl) {
1873                 struct page *page;
1874
1875                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1876                 if (page)
1877                         gather_stats(page, md, pmd_dirty(*pmd),
1878                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1879                 spin_unlock(ptl);
1880                 return 0;
1881         }
1882 #endif
1883         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1884         if (!pte) {
1885                 walk->action = ACTION_AGAIN;
1886                 return 0;
1887         }
1888         do {
1889                 pte_t ptent = ptep_get(pte);
1890                 struct page *page = can_gather_numa_stats(ptent, vma, addr);
1891                 if (!page)
1892                         continue;
1893                 gather_stats(page, md, pte_dirty(ptent), 1);
1894
1895         } while (pte++, addr += PAGE_SIZE, addr != end);
1896         pte_unmap_unlock(orig_pte, ptl);
1897         cond_resched();
1898         return 0;
1899 }
1900 #ifdef CONFIG_HUGETLB_PAGE
1901 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1902                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1903 {
1904         pte_t huge_pte = huge_ptep_get(pte);
1905         struct numa_maps *md;
1906         struct page *page;
1907
1908         if (!pte_present(huge_pte))
1909                 return 0;
1910
1911         page = pte_page(huge_pte);
1912
1913         md = walk->private;
1914         gather_stats(page, md, pte_dirty(huge_pte), 1);
1915         return 0;
1916 }
1917
1918 #else
1919 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1920                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1921 {
1922         return 0;
1923 }
1924 #endif
1925
1926 static const struct mm_walk_ops show_numa_ops = {
1927         .hugetlb_entry = gather_hugetlb_stats,
1928         .pmd_entry = gather_pte_stats,
1929         .walk_lock = PGWALK_RDLOCK,
1930 };
1931
1932 /*
1933  * Display pages allocated per node and memory policy via /proc.
1934  */
1935 static int show_numa_map(struct seq_file *m, void *v)
1936 {
1937         struct numa_maps_private *numa_priv = m->private;
1938         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1939         struct vm_area_struct *vma = v;
1940         struct numa_maps *md = &numa_priv->md;
1941         struct file *file = vma->vm_file;
1942         struct mm_struct *mm = vma->vm_mm;
1943         struct mempolicy *pol;
1944         char buffer[64];
1945         int nid;
1946
1947         if (!mm)
1948                 return 0;
1949
1950         /* Ensure we start with an empty set of numa_maps statistics. */
1951         memset(md, 0, sizeof(*md));
1952
1953         pol = __get_vma_policy(vma, vma->vm_start);
1954         if (pol) {
1955                 mpol_to_str(buffer, sizeof(buffer), pol);
1956                 mpol_cond_put(pol);
1957         } else {
1958                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1959         }
1960
1961         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1962
1963         if (file) {
1964                 seq_puts(m, " file=");
1965                 seq_file_path(m, file, "\n\t= ");
1966         } else if (vma_is_initial_heap(vma)) {
1967                 seq_puts(m, " heap");
1968         } else if (vma_is_initial_stack(vma)) {
1969                 seq_puts(m, " stack");
1970         }
1971
1972         if (is_vm_hugetlb_page(vma))
1973                 seq_puts(m, " huge");
1974
1975         /* mmap_lock is held by m_start */
1976         walk_page_vma(vma, &show_numa_ops, md);
1977
1978         if (!md->pages)
1979                 goto out;
1980
1981         if (md->anon)
1982                 seq_printf(m, " anon=%lu", md->anon);
1983
1984         if (md->dirty)
1985                 seq_printf(m, " dirty=%lu", md->dirty);
1986
1987         if (md->pages != md->anon && md->pages != md->dirty)
1988                 seq_printf(m, " mapped=%lu", md->pages);
1989
1990         if (md->mapcount_max > 1)
1991                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1992
1993         if (md->swapcache)
1994                 seq_printf(m, " swapcache=%lu", md->swapcache);
1995
1996         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1997                 seq_printf(m, " active=%lu", md->active);
1998
1999         if (md->writeback)
2000                 seq_printf(m, " writeback=%lu", md->writeback);
2001
2002         for_each_node_state(nid, N_MEMORY)
2003                 if (md->node[nid])
2004                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2005
2006         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2007 out:
2008         seq_putc(m, '\n');
2009         return 0;
2010 }
2011
2012 static const struct seq_operations proc_pid_numa_maps_op = {
2013         .start  = m_start,
2014         .next   = m_next,
2015         .stop   = m_stop,
2016         .show   = show_numa_map,
2017 };
2018
2019 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2020 {
2021         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2022                                 sizeof(struct numa_maps_private));
2023 }
2024
2025 const struct file_operations proc_pid_numa_maps_operations = {
2026         .open           = pid_numa_maps_open,
2027         .read           = seq_read,
2028         .llseek         = seq_lseek,
2029         .release        = proc_map_release,
2030 };
2031
2032 #endif /* CONFIG_NUMA */