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         memset(&mss, 0, sizeof(mss));
 855
 856         smap_gather_stats(vma, &mss, 0);
 857
 858         show_map_vma(m, vma);
 859
 860         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 861         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 862         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 863         seq_puts(m, " kB\n");
 864
 865         __show_smap(m, &mss, false);
 866
 867         seq_printf(m, "THPeligible:    %8u\n",
 868                    hugepage_vma_check(vma, vma->vm_flags, true, false, true));
 869
 870         if (arch_pkeys_enabled())
 871                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 872         show_smap_vma_flags(m, vma);
 873
 874         return 0;
 875 }
 876
 877 static int show_smaps_rollup(struct seq_file *m, void *v)
 878 {
 879         struct proc_maps_private *priv = m->private;
 880         struct mem_size_stats mss;
 881         struct mm_struct *mm = priv->mm;
 882         struct vm_area_struct *vma;
 883         unsigned long vma_start = 0, last_vma_end = 0;
 884         int ret = 0;
 885         VMA_ITERATOR(vmi, mm, 0);
 886
 887         priv->task = get_proc_task(priv->inode);
 888         if (!priv->task)
 889                 return -ESRCH;
 890
 891         if (!mm || !mmget_not_zero(mm)) {
 892                 ret = -ESRCH;
 893                 goto out_put_task;
 894         }
 895
 896         memset(&mss, 0, sizeof(mss));
 897
 898         ret = mmap_read_lock_killable(mm);
 899         if (ret)
 900                 goto out_put_mm;
 901
 902         hold_task_mempolicy(priv);
 903         vma = vma_next(&vmi);
 904
 905         if (unlikely(!vma))
 906                 goto empty_set;
 907
 908         vma_start = vma->vm_start;
 909         do {
 910                 smap_gather_stats(vma, &mss, 0);
 911                 last_vma_end = vma->vm_end;
 912
 913                 /*
 914                  * Release mmap_lock temporarily if someone wants to
 915                  * access it for write request.
 916                  */
 917                 if (mmap_lock_is_contended(mm)) {
 918                         vma_iter_invalidate(&vmi);
 919                         mmap_read_unlock(mm);
 920                         ret = mmap_read_lock_killable(mm);
 921                         if (ret) {
 922                                 release_task_mempolicy(priv);
 923                                 goto out_put_mm;
 924                         }
 925
 926                         /*
 927                          * After dropping the lock, there are four cases to
 928                          * consider. See the following example for explanation.
 929                          *
 930                          *   +------+------+-----------+
 931                          *   | VMA1 | VMA2 | VMA3      |
 932                          *   +------+------+-----------+
 933                          *   |      |      |           |
 934                          *  4k     8k     16k         400k
 935                          *
 936                          * Suppose we drop the lock after reading VMA2 due to
 937                          * contention, then we get:
 938                          *
 939                          *      last_vma_end = 16k
 940                          *
 941                          * 1) VMA2 is freed, but VMA3 exists:
 942                          *
 943                          *    vma_next(vmi) will return VMA3.
 944                          *    In this case, just continue from VMA3.
 945                          *
 946                          * 2) VMA2 still exists:
 947                          *
 948                          *    vma_next(vmi) will return VMA3.
 949                          *    In this case, just continue from VMA3.
 950                          *
 951                          * 3) No more VMAs can be found:
 952                          *
 953                          *    vma_next(vmi) will return NULL.
 954                          *    No more things to do, just break.
 955                          *
 956                          * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
 957                          *
 958                          *    vma_next(vmi) will return VMA' whose range
 959                          *    contains last_vma_end.
 960                          *    Iterate VMA' from last_vma_end.
 961                          */
 962                         vma = vma_next(&vmi);
 963                         /* Case 3 above */
 964                         if (!vma)
 965                                 break;
 966
 967                         /* Case 1 and 2 above */
 968                         if (vma->vm_start >= last_vma_end)
 969                                 continue;
 970
 971                         /* Case 4 above */
 972                         if (vma->vm_end > last_vma_end)
 973                                 smap_gather_stats(vma, &mss, last_vma_end);
 974                 }
 975         } for_each_vma(vmi, vma);
 976
 977 empty_set:
 978         show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
 979         seq_pad(m, ' ');
 980         seq_puts(m, "[rollup]\n");
 981
 982         __show_smap(m, &mss, true);
 983
 984         release_task_mempolicy(priv);
 985         mmap_read_unlock(mm);
 986
 987 out_put_mm:
 988         mmput(mm);
 989 out_put_task:
 990         put_task_struct(priv->task);
 991         priv->task = NULL;
 992
 993         return ret;
 994 }
 995 #undef SEQ_PUT_DEC
 996
 997 static const struct seq_operations proc_pid_smaps_op = {
 998         .start  = m_start,
 999         .next   = m_next,
1000         .stop   = m_stop,
1001         .show   = show_smap
1002 };
1003
1004 static int pid_smaps_open(struct inode *inode, struct file *file)
1005 {
1006         return do_maps_open(inode, file, &proc_pid_smaps_op);
1007 }
1008
1009 static int smaps_rollup_open(struct inode *inode, struct file *file)
1010 {
1011         int ret;
1012         struct proc_maps_private *priv;
1013
1014         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1015         if (!priv)
1016                 return -ENOMEM;
1017
1018         ret = single_open(file, show_smaps_rollup, priv);
1019         if (ret)
1020                 goto out_free;
1021
1022         priv->inode = inode;
1023         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1024         if (IS_ERR(priv->mm)) {
1025                 ret = PTR_ERR(priv->mm);
1026
1027                 single_release(inode, file);
1028                 goto out_free;
1029         }
1030
1031         return 0;
1032
1033 out_free:
1034         kfree(priv);
1035         return ret;
1036 }
1037
1038 static int smaps_rollup_release(struct inode *inode, struct file *file)
1039 {
1040         struct seq_file *seq = file->private_data;
1041         struct proc_maps_private *priv = seq->private;
1042
1043         if (priv->mm)
1044                 mmdrop(priv->mm);
1045
1046         kfree(priv);
1047         return single_release(inode, file);
1048 }
1049
1050 const struct file_operations proc_pid_smaps_operations = {
1051         .open           = pid_smaps_open,
1052         .read           = seq_read,
1053         .llseek         = seq_lseek,
1054         .release        = proc_map_release,
1055 };
1056
1057 const struct file_operations proc_pid_smaps_rollup_operations = {
1058         .open           = smaps_rollup_open,
1059         .read           = seq_read,
1060         .llseek         = seq_lseek,
1061         .release        = smaps_rollup_release,
1062 };
1063
1064 enum clear_refs_types {
1065         CLEAR_REFS_ALL = 1,
1066         CLEAR_REFS_ANON,
1067         CLEAR_REFS_MAPPED,
1068         CLEAR_REFS_SOFT_DIRTY,
1069         CLEAR_REFS_MM_HIWATER_RSS,
1070         CLEAR_REFS_LAST,
1071 };
1072
1073 struct clear_refs_private {
1074         enum clear_refs_types type;
1075 };
1076
1077 #ifdef CONFIG_MEM_SOFT_DIRTY
1078
1079 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1080 {
1081         struct page *page;
1082
1083         if (!pte_write(pte))
1084                 return false;
1085         if (!is_cow_mapping(vma->vm_flags))
1086                 return false;
1087         if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1088                 return false;
1089         page = vm_normal_page(vma, addr, pte);
1090         if (!page)
1091                 return false;
1092         return page_maybe_dma_pinned(page);
1093 }
1094
1095 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1096                 unsigned long addr, pte_t *pte)
1097 {
1098         /*
1099          * The soft-dirty tracker uses #PF-s to catch writes
1100          * to pages, so write-protect the pte as well. See the
1101          * Documentation/admin-guide/mm/soft-dirty.rst for full description
1102          * of how soft-dirty works.
1103          */
1104         pte_t ptent = ptep_get(pte);
1105
1106         if (pte_present(ptent)) {
1107                 pte_t old_pte;
1108
1109                 if (pte_is_pinned(vma, addr, ptent))
1110                         return;
1111                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1112                 ptent = pte_wrprotect(old_pte);
1113                 ptent = pte_clear_soft_dirty(ptent);
1114                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1115         } else if (is_swap_pte(ptent)) {
1116                 ptent = pte_swp_clear_soft_dirty(ptent);
1117                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1118         }
1119 }
1120 #else
1121 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1122                 unsigned long addr, pte_t *pte)
1123 {
1124 }
1125 #endif
1126
1127 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1128 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1129                 unsigned long addr, pmd_t *pmdp)
1130 {
1131         pmd_t old, pmd = *pmdp;
1132
1133         if (pmd_present(pmd)) {
1134                 /* See comment in change_huge_pmd() */
1135                 old = pmdp_invalidate(vma, addr, pmdp);
1136                 if (pmd_dirty(old))
1137                         pmd = pmd_mkdirty(pmd);
1138                 if (pmd_young(old))
1139                         pmd = pmd_mkyoung(pmd);
1140
1141                 pmd = pmd_wrprotect(pmd);
1142                 pmd = pmd_clear_soft_dirty(pmd);
1143
1144                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1145         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1146                 pmd = pmd_swp_clear_soft_dirty(pmd);
1147                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1148         }
1149 }
1150 #else
1151 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1152                 unsigned long addr, pmd_t *pmdp)
1153 {
1154 }
1155 #endif
1156
1157 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1158                                 unsigned long end, struct mm_walk *walk)
1159 {
1160         struct clear_refs_private *cp = walk->private;
1161         struct vm_area_struct *vma = walk->vma;
1162         pte_t *pte, ptent;
1163         spinlock_t *ptl;
1164         struct page *page;
1165
1166         ptl = pmd_trans_huge_lock(pmd, vma);
1167         if (ptl) {
1168                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1169                         clear_soft_dirty_pmd(vma, addr, pmd);
1170                         goto out;
1171                 }
1172
1173                 if (!pmd_present(*pmd))
1174                         goto out;
1175
1176                 page = pmd_page(*pmd);
1177
1178                 /* Clear accessed and referenced bits. */
1179                 pmdp_test_and_clear_young(vma, addr, pmd);
1180                 test_and_clear_page_young(page);
1181                 ClearPageReferenced(page);
1182 out:
1183                 spin_unlock(ptl);
1184                 return 0;
1185         }
1186
1187         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1188         if (!pte) {
1189                 walk->action = ACTION_AGAIN;
1190                 return 0;
1191         }
1192         for (; addr != end; pte++, addr += PAGE_SIZE) {
1193                 ptent = ptep_get(pte);
1194
1195                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1196                         clear_soft_dirty(vma, addr, pte);
1197                         continue;
1198                 }
1199
1200                 if (!pte_present(ptent))
1201                         continue;
1202
1203                 page = vm_normal_page(vma, addr, ptent);
1204                 if (!page)
1205                         continue;
1206
1207                 /* Clear accessed and referenced bits. */
1208                 ptep_test_and_clear_young(vma, addr, pte);
1209                 test_and_clear_page_young(page);
1210                 ClearPageReferenced(page);
1211         }
1212         pte_unmap_unlock(pte - 1, ptl);
1213         cond_resched();
1214         return 0;
1215 }
1216
1217 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1218                                 struct mm_walk *walk)
1219 {
1220         struct clear_refs_private *cp = walk->private;
1221         struct vm_area_struct *vma = walk->vma;
1222
1223         if (vma->vm_flags & VM_PFNMAP)
1224                 return 1;
1225
1226         /*
1227          * Writing 1 to /proc/pid/clear_refs affects all pages.
1228          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1229          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1230          * Writing 4 to /proc/pid/clear_refs affects all pages.
1231          */
1232         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1233                 return 1;
1234         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1235                 return 1;
1236         return 0;
1237 }
1238
1239 static const struct mm_walk_ops clear_refs_walk_ops = {
1240         .pmd_entry              = clear_refs_pte_range,
1241         .test_walk              = clear_refs_test_walk,
1242         .walk_lock              = PGWALK_WRLOCK,
1243 };
1244
1245 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1246                                 size_t count, loff_t *ppos)
1247 {
1248         struct task_struct *task;
1249         char buffer[PROC_NUMBUF];
1250         struct mm_struct *mm;
1251         struct vm_area_struct *vma;
1252         enum clear_refs_types type;
1253         int itype;
1254         int rv;
1255
1256         memset(buffer, 0, sizeof(buffer));
1257         if (count > sizeof(buffer) - 1)
1258                 count = sizeof(buffer) - 1;
1259         if (copy_from_user(buffer, buf, count))
1260                 return -EFAULT;
1261         rv = kstrtoint(strstrip(buffer), 10, &itype);
1262         if (rv < 0)
1263                 return rv;
1264         type = (enum clear_refs_types)itype;
1265         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1266                 return -EINVAL;
1267
1268         task = get_proc_task(file_inode(file));
1269         if (!task)
1270                 return -ESRCH;
1271         mm = get_task_mm(task);
1272         if (mm) {
1273                 VMA_ITERATOR(vmi, mm, 0);
1274                 struct mmu_notifier_range range;
1275                 struct clear_refs_private cp = {
1276                         .type = type,
1277                 };
1278
1279                 if (mmap_write_lock_killable(mm)) {
1280                         count = -EINTR;
1281                         goto out_mm;
1282                 }
1283                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1284                         /*
1285                          * Writing 5 to /proc/pid/clear_refs resets the peak
1286                          * resident set size to this mm's current rss value.
1287                          */
1288                         reset_mm_hiwater_rss(mm);
1289                         goto out_unlock;
1290                 }
1291
1292                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1293                         for_each_vma(vmi, vma) {
1294                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1295                                         continue;
1296                                 vm_flags_clear(vma, VM_SOFTDIRTY);
1297                                 vma_set_page_prot(vma);
1298                         }
1299
1300                         inc_tlb_flush_pending(mm);
1301                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1302                                                 0, mm, 0, -1UL);
1303                         mmu_notifier_invalidate_range_start(&range);
1304                 }
1305                 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1306                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1307                         mmu_notifier_invalidate_range_end(&range);
1308                         flush_tlb_mm(mm);
1309                         dec_tlb_flush_pending(mm);
1310                 }
1311 out_unlock:
1312                 mmap_write_unlock(mm);
1313 out_mm:
1314                 mmput(mm);
1315         }
1316         put_task_struct(task);
1317
1318         return count;
1319 }
1320
1321 const struct file_operations proc_clear_refs_operations = {
1322         .write          = clear_refs_write,
1323         .llseek         = noop_llseek,
1324 };
1325
1326 typedef struct {
1327         u64 pme;
1328 } pagemap_entry_t;
1329
1330 struct pagemapread {
1331         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1332         pagemap_entry_t *buffer;
1333         bool show_pfn;
1334 };
1335
1336 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1337 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1338
1339 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1340 #define PM_PFRAME_BITS          55
1341 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1342 #define PM_SOFT_DIRTY           BIT_ULL(55)
1343 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1344 #define PM_UFFD_WP              BIT_ULL(57)
1345 #define PM_FILE                 BIT_ULL(61)
1346 #define PM_SWAP                 BIT_ULL(62)
1347 #define PM_PRESENT              BIT_ULL(63)
1348
1349 #define PM_END_OF_BUFFER    1
1350
1351 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1352 {
1353         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1354 }
1355
1356 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1357                           struct pagemapread *pm)
1358 {
1359         pm->buffer[pm->pos++] = *pme;
1360         if (pm->pos >= pm->len)
1361                 return PM_END_OF_BUFFER;
1362         return 0;
1363 }
1364
1365 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1366                             __always_unused int depth, struct mm_walk *walk)
1367 {
1368         struct pagemapread *pm = walk->private;
1369         unsigned long addr = start;
1370         int err = 0;
1371
1372         while (addr < end) {
1373                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1374                 pagemap_entry_t pme = make_pme(0, 0);
1375                 /* End of address space hole, which we mark as non-present. */
1376                 unsigned long hole_end;
1377
1378                 if (vma)
1379                         hole_end = min(end, vma->vm_start);
1380                 else
1381                         hole_end = end;
1382
1383                 for (; addr < hole_end; addr += PAGE_SIZE) {
1384                         err = add_to_pagemap(addr, &pme, pm);
1385                         if (err)
1386                                 goto out;
1387                 }
1388
1389                 if (!vma)
1390                         break;
1391
1392                 /* Addresses in the VMA. */
1393                 if (vma->vm_flags & VM_SOFTDIRTY)
1394                         pme = make_pme(0, PM_SOFT_DIRTY);
1395                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1396                         err = add_to_pagemap(addr, &pme, pm);
1397                         if (err)
1398                                 goto out;
1399                 }
1400         }
1401 out:
1402         return err;
1403 }
1404
1405 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1406                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1407 {
1408         u64 frame = 0, flags = 0;
1409         struct page *page = NULL;
1410         bool migration = false;
1411
1412         if (pte_present(pte)) {
1413                 if (pm->show_pfn)
1414                         frame = pte_pfn(pte);
1415                 flags |= PM_PRESENT;
1416                 page = vm_normal_page(vma, addr, pte);
1417                 if (pte_soft_dirty(pte))
1418                         flags |= PM_SOFT_DIRTY;
1419                 if (pte_uffd_wp(pte))
1420                         flags |= PM_UFFD_WP;
1421         } else if (is_swap_pte(pte)) {
1422                 swp_entry_t entry;
1423                 if (pte_swp_soft_dirty(pte))
1424                         flags |= PM_SOFT_DIRTY;
1425                 if (pte_swp_uffd_wp(pte))
1426                         flags |= PM_UFFD_WP;
1427                 entry = pte_to_swp_entry(pte);
1428                 if (pm->show_pfn) {
1429                         pgoff_t offset;
1430                         /*
1431                          * For PFN swap offsets, keeping the offset field
1432                          * to be PFN only to be compatible with old smaps.
1433                          */
1434                         if (is_pfn_swap_entry(entry))
1435                                 offset = swp_offset_pfn(entry);
1436                         else
1437                                 offset = swp_offset(entry);
1438                         frame = swp_type(entry) |
1439                             (offset << MAX_SWAPFILES_SHIFT);
1440                 }
1441                 flags |= PM_SWAP;
1442                 migration = is_migration_entry(entry);
1443                 if (is_pfn_swap_entry(entry))
1444                         page = pfn_swap_entry_to_page(entry);
1445                 if (pte_marker_entry_uffd_wp(entry))
1446                         flags |= PM_UFFD_WP;
1447         }
1448
1449         if (page && !PageAnon(page))
1450                 flags |= PM_FILE;
1451         if (page && !migration && page_mapcount(page) == 1)
1452                 flags |= PM_MMAP_EXCLUSIVE;
1453         if (vma->vm_flags & VM_SOFTDIRTY)
1454                 flags |= PM_SOFT_DIRTY;
1455
1456         return make_pme(frame, flags);
1457 }
1458
1459 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1460                              struct mm_walk *walk)
1461 {
1462         struct vm_area_struct *vma = walk->vma;
1463         struct pagemapread *pm = walk->private;
1464         spinlock_t *ptl;
1465         pte_t *pte, *orig_pte;
1466         int err = 0;
1467 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1468         bool migration = false;
1469
1470         ptl = pmd_trans_huge_lock(pmdp, vma);
1471         if (ptl) {
1472                 u64 flags = 0, frame = 0;
1473                 pmd_t pmd = *pmdp;
1474                 struct page *page = NULL;
1475
1476                 if (vma->vm_flags & VM_SOFTDIRTY)
1477                         flags |= PM_SOFT_DIRTY;
1478
1479                 if (pmd_present(pmd)) {
1480                         page = pmd_page(pmd);
1481
1482                         flags |= PM_PRESENT;
1483                         if (pmd_soft_dirty(pmd))
1484                                 flags |= PM_SOFT_DIRTY;
1485                         if (pmd_uffd_wp(pmd))
1486                                 flags |= PM_UFFD_WP;
1487                         if (pm->show_pfn)
1488                                 frame = pmd_pfn(pmd) +
1489                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1490                 }
1491 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1492                 else if (is_swap_pmd(pmd)) {
1493                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1494                         unsigned long offset;
1495
1496                         if (pm->show_pfn) {
1497                                 if (is_pfn_swap_entry(entry))
1498                                         offset = swp_offset_pfn(entry);
1499                                 else
1500                                         offset = swp_offset(entry);
1501                                 offset = offset +
1502                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1503                                 frame = swp_type(entry) |
1504                                         (offset << MAX_SWAPFILES_SHIFT);
1505                         }
1506                         flags |= PM_SWAP;
1507                         if (pmd_swp_soft_dirty(pmd))
1508                                 flags |= PM_SOFT_DIRTY;
1509                         if (pmd_swp_uffd_wp(pmd))
1510                                 flags |= PM_UFFD_WP;
1511                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1512                         migration = is_migration_entry(entry);
1513                         page = pfn_swap_entry_to_page(entry);
1514                 }
1515 #endif
1516
1517                 if (page && !migration && page_mapcount(page) == 1)
1518                         flags |= PM_MMAP_EXCLUSIVE;
1519
1520                 for (; addr != end; addr += PAGE_SIZE) {
1521                         pagemap_entry_t pme = make_pme(frame, flags);
1522
1523                         err = add_to_pagemap(addr, &pme, pm);
1524                         if (err)
1525                                 break;
1526                         if (pm->show_pfn) {
1527                                 if (flags & PM_PRESENT)
1528                                         frame++;
1529                                 else if (flags & PM_SWAP)
1530                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1531                         }
1532                 }
1533                 spin_unlock(ptl);
1534                 return err;
1535         }
1536 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1537
1538         /*
1539          * We can assume that @vma always points to a valid one and @end never
1540          * goes beyond vma->vm_end.
1541          */
1542         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1543         if (!pte) {
1544                 walk->action = ACTION_AGAIN;
1545                 return err;
1546         }
1547         for (; addr < end; pte++, addr += PAGE_SIZE) {
1548                 pagemap_entry_t pme;
1549
1550                 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte));
1551                 err = add_to_pagemap(addr, &pme, pm);
1552                 if (err)
1553                         break;
1554         }
1555         pte_unmap_unlock(orig_pte, ptl);
1556
1557         cond_resched();
1558
1559         return err;
1560 }
1561
1562 #ifdef CONFIG_HUGETLB_PAGE
1563 /* This function walks within one hugetlb entry in the single call */
1564 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1565                                  unsigned long addr, unsigned long end,
1566                                  struct mm_walk *walk)
1567 {
1568         struct pagemapread *pm = walk->private;
1569         struct vm_area_struct *vma = walk->vma;
1570         u64 flags = 0, frame = 0;
1571         int err = 0;
1572         pte_t pte;
1573
1574         if (vma->vm_flags & VM_SOFTDIRTY)
1575                 flags |= PM_SOFT_DIRTY;
1576
1577         pte = huge_ptep_get(ptep);
1578         if (pte_present(pte)) {
1579                 struct page *page = pte_page(pte);
1580
1581                 if (!PageAnon(page))
1582                         flags |= PM_FILE;
1583
1584                 if (page_mapcount(page) == 1)
1585                         flags |= PM_MMAP_EXCLUSIVE;
1586
1587                 if (huge_pte_uffd_wp(pte))
1588                         flags |= PM_UFFD_WP;
1589
1590                 flags |= PM_PRESENT;
1591                 if (pm->show_pfn)
1592                         frame = pte_pfn(pte) +
1593                                 ((addr & ~hmask) >> PAGE_SHIFT);
1594         } else if (pte_swp_uffd_wp_any(pte)) {
1595                 flags |= PM_UFFD_WP;
1596         }
1597
1598         for (; addr != end; addr += PAGE_SIZE) {
1599                 pagemap_entry_t pme = make_pme(frame, flags);
1600
1601                 err = add_to_pagemap(addr, &pme, pm);
1602                 if (err)
1603                         return err;
1604                 if (pm->show_pfn && (flags & PM_PRESENT))
1605                         frame++;
1606         }
1607
1608         cond_resched();
1609
1610         return err;
1611 }
1612 #else
1613 #define pagemap_hugetlb_range   NULL
1614 #endif /* HUGETLB_PAGE */
1615
1616 static const struct mm_walk_ops pagemap_ops = {
1617         .pmd_entry      = pagemap_pmd_range,
1618         .pte_hole       = pagemap_pte_hole,
1619         .hugetlb_entry  = pagemap_hugetlb_range,
1620         .walk_lock      = PGWALK_RDLOCK,
1621 };
1622
1623 /*
1624  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1625  *
1626  * For each page in the address space, this file contains one 64-bit entry
1627  * consisting of the following:
1628  *
1629  * Bits 0-54  page frame number (PFN) if present
1630  * Bits 0-4   swap type if swapped
1631  * Bits 5-54  swap offset if swapped
1632  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1633  * Bit  56    page exclusively mapped
1634  * Bit  57    pte is uffd-wp write-protected
1635  * Bits 58-60 zero
1636  * Bit  61    page is file-page or shared-anon
1637  * Bit  62    page swapped
1638  * Bit  63    page present
1639  *
1640  * If the page is not present but in swap, then the PFN contains an
1641  * encoding of the swap file number and the page's offset into the
1642  * swap. Unmapped pages return a null PFN. This allows determining
1643  * precisely which pages are mapped (or in swap) and comparing mapped
1644  * pages between processes.
1645  *
1646  * Efficient users of this interface will use /proc/pid/maps to
1647  * determine which areas of memory are actually mapped and llseek to
1648  * skip over unmapped regions.
1649  */
1650 static ssize_t pagemap_read(struct file *file, char __user *buf,
1651                             size_t count, loff_t *ppos)
1652 {
1653         struct mm_struct *mm = file->private_data;
1654         struct pagemapread pm;
1655         unsigned long src;
1656         unsigned long svpfn;
1657         unsigned long start_vaddr;
1658         unsigned long end_vaddr;
1659         int ret = 0, copied = 0;
1660
1661         if (!mm || !mmget_not_zero(mm))
1662                 goto out;
1663
1664         ret = -EINVAL;
1665         /* file position must be aligned */
1666         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1667                 goto out_mm;
1668
1669         ret = 0;
1670         if (!count)
1671                 goto out_mm;
1672
1673         /* do not disclose physical addresses: attack vector */
1674         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1675
1676         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1677         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1678         ret = -ENOMEM;
1679         if (!pm.buffer)
1680                 goto out_mm;
1681
1682         src = *ppos;
1683         svpfn = src / PM_ENTRY_BYTES;
1684         end_vaddr = mm->task_size;
1685
1686         /* watch out for wraparound */
1687         start_vaddr = end_vaddr;
1688         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
1689                 unsigned long end;
1690
1691                 ret = mmap_read_lock_killable(mm);
1692                 if (ret)
1693                         goto out_free;
1694                 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
1695                 mmap_read_unlock(mm);
1696
1697                 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
1698                 if (end >= start_vaddr && end < mm->task_size)
1699                         end_vaddr = end;
1700         }
1701
1702         /* Ensure the address is inside the task */
1703         if (start_vaddr > mm->task_size)
1704                 start_vaddr = end_vaddr;
1705
1706         ret = 0;
1707         while (count && (start_vaddr < end_vaddr)) {
1708                 int len;
1709                 unsigned long end;
1710
1711                 pm.pos = 0;
1712                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1713                 /* overflow ? */
1714                 if (end < start_vaddr || end > end_vaddr)
1715                         end = end_vaddr;
1716                 ret = mmap_read_lock_killable(mm);
1717                 if (ret)
1718                         goto out_free;
1719                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1720                 mmap_read_unlock(mm);
1721                 start_vaddr = end;
1722
1723                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1724                 if (copy_to_user(buf, pm.buffer, len)) {
1725                         ret = -EFAULT;
1726                         goto out_free;
1727                 }
1728                 copied += len;
1729                 buf += len;
1730                 count -= len;
1731         }
1732         *ppos += copied;
1733         if (!ret || ret == PM_END_OF_BUFFER)
1734                 ret = copied;
1735
1736 out_free:
1737         kfree(pm.buffer);
1738 out_mm:
1739         mmput(mm);
1740 out:
1741         return ret;
1742 }
1743
1744 static int pagemap_open(struct inode *inode, struct file *file)
1745 {
1746         struct mm_struct *mm;
1747
1748         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1749         if (IS_ERR(mm))
1750                 return PTR_ERR(mm);
1751         file->private_data = mm;
1752         return 0;
1753 }
1754
1755 static int pagemap_release(struct inode *inode, struct file *file)
1756 {
1757         struct mm_struct *mm = file->private_data;
1758
1759         if (mm)
1760                 mmdrop(mm);
1761         return 0;
1762 }
1763
1764 const struct file_operations proc_pagemap_operations = {
1765         .llseek         = mem_lseek, /* borrow this */
1766         .read           = pagemap_read,
1767         .open           = pagemap_open,
1768         .release        = pagemap_release,
1769 };
1770 #endif /* CONFIG_PROC_PAGE_MONITOR */
1771
1772 #ifdef CONFIG_NUMA
1773
1774 struct numa_maps {
1775         unsigned long pages;
1776         unsigned long anon;
1777         unsigned long active;
1778         unsigned long writeback;
1779         unsigned long mapcount_max;
1780         unsigned long dirty;
1781         unsigned long swapcache;
1782         unsigned long node[MAX_NUMNODES];
1783 };
1784
1785 struct numa_maps_private {
1786         struct proc_maps_private proc_maps;
1787         struct numa_maps md;
1788 };
1789
1790 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1791                         unsigned long nr_pages)
1792 {
1793         int count = page_mapcount(page);
1794
1795         md->pages += nr_pages;
1796         if (pte_dirty || PageDirty(page))
1797                 md->dirty += nr_pages;
1798
1799         if (PageSwapCache(page))
1800                 md->swapcache += nr_pages;
1801
1802         if (PageActive(page) || PageUnevictable(page))
1803                 md->active += nr_pages;
1804
1805         if (PageWriteback(page))
1806                 md->writeback += nr_pages;
1807
1808         if (PageAnon(page))
1809                 md->anon += nr_pages;
1810
1811         if (count > md->mapcount_max)
1812                 md->mapcount_max = count;
1813
1814         md->node[page_to_nid(page)] += nr_pages;
1815 }
1816
1817 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1818                 unsigned long addr)
1819 {
1820         struct page *page;
1821         int nid;
1822
1823         if (!pte_present(pte))
1824                 return NULL;
1825
1826         page = vm_normal_page(vma, addr, pte);
1827         if (!page || is_zone_device_page(page))
1828                 return NULL;
1829
1830         if (PageReserved(page))
1831                 return NULL;
1832
1833         nid = page_to_nid(page);
1834         if (!node_isset(nid, node_states[N_MEMORY]))
1835                 return NULL;
1836
1837         return page;
1838 }
1839
1840 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1841 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1842                                               struct vm_area_struct *vma,
1843                                               unsigned long addr)
1844 {
1845         struct page *page;
1846         int nid;
1847
1848         if (!pmd_present(pmd))
1849                 return NULL;
1850
1851         page = vm_normal_page_pmd(vma, addr, pmd);
1852         if (!page)
1853                 return NULL;
1854
1855         if (PageReserved(page))
1856                 return NULL;
1857
1858         nid = page_to_nid(page);
1859         if (!node_isset(nid, node_states[N_MEMORY]))
1860                 return NULL;
1861
1862         return page;
1863 }
1864 #endif
1865
1866 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1867                 unsigned long end, struct mm_walk *walk)
1868 {
1869         struct numa_maps *md = walk->private;
1870         struct vm_area_struct *vma = walk->vma;
1871         spinlock_t *ptl;
1872         pte_t *orig_pte;
1873         pte_t *pte;
1874
1875 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1876         ptl = pmd_trans_huge_lock(pmd, vma);
1877         if (ptl) {
1878                 struct page *page;
1879
1880                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1881                 if (page)
1882                         gather_stats(page, md, pmd_dirty(*pmd),
1883                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1884                 spin_unlock(ptl);
1885                 return 0;
1886         }
1887 #endif
1888         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1889         if (!pte) {
1890                 walk->action = ACTION_AGAIN;
1891                 return 0;
1892         }
1893         do {
1894                 pte_t ptent = ptep_get(pte);
1895                 struct page *page = can_gather_numa_stats(ptent, vma, addr);
1896                 if (!page)
1897                         continue;
1898                 gather_stats(page, md, pte_dirty(ptent), 1);
1899
1900         } while (pte++, addr += PAGE_SIZE, addr != end);
1901         pte_unmap_unlock(orig_pte, ptl);
1902         cond_resched();
1903         return 0;
1904 }
1905 #ifdef CONFIG_HUGETLB_PAGE
1906 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1907                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1908 {
1909         pte_t huge_pte = huge_ptep_get(pte);
1910         struct numa_maps *md;
1911         struct page *page;
1912
1913         if (!pte_present(huge_pte))
1914                 return 0;
1915
1916         page = pte_page(huge_pte);
1917
1918         md = walk->private;
1919         gather_stats(page, md, pte_dirty(huge_pte), 1);
1920         return 0;
1921 }
1922
1923 #else
1924 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1925                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1926 {
1927         return 0;
1928 }
1929 #endif
1930
1931 static const struct mm_walk_ops show_numa_ops = {
1932         .hugetlb_entry = gather_hugetlb_stats,
1933         .pmd_entry = gather_pte_stats,
1934         .walk_lock = PGWALK_RDLOCK,
1935 };
1936
1937 /*
1938  * Display pages allocated per node and memory policy via /proc.
1939  */
1940 static int show_numa_map(struct seq_file *m, void *v)
1941 {
1942         struct numa_maps_private *numa_priv = m->private;
1943         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1944         struct vm_area_struct *vma = v;
1945         struct numa_maps *md = &numa_priv->md;
1946         struct file *file = vma->vm_file;
1947         struct mm_struct *mm = vma->vm_mm;
1948         struct mempolicy *pol;
1949         char buffer[64];
1950         int nid;
1951
1952         if (!mm)
1953                 return 0;
1954
1955         /* Ensure we start with an empty set of numa_maps statistics. */
1956         memset(md, 0, sizeof(*md));
1957
1958         pol = __get_vma_policy(vma, vma->vm_start);
1959         if (pol) {
1960                 mpol_to_str(buffer, sizeof(buffer), pol);
1961                 mpol_cond_put(pol);
1962         } else {
1963                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1964         }
1965
1966         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1967
1968         if (file) {
1969                 seq_puts(m, " file=");
1970                 seq_file_path(m, file, "\n\t= ");
1971         } else if (vma_is_initial_heap(vma)) {
1972                 seq_puts(m, " heap");
1973         } else if (vma_is_initial_stack(vma)) {
1974                 seq_puts(m, " stack");
1975         }
1976
1977         if (is_vm_hugetlb_page(vma))
1978                 seq_puts(m, " huge");
1979
1980         /* mmap_lock is held by m_start */
1981         walk_page_vma(vma, &show_numa_ops, md);
1982
1983         if (!md->pages)
1984                 goto out;
1985
1986         if (md->anon)
1987                 seq_printf(m, " anon=%lu", md->anon);
1988
1989         if (md->dirty)
1990                 seq_printf(m, " dirty=%lu", md->dirty);
1991
1992         if (md->pages != md->anon && md->pages != md->dirty)
1993                 seq_printf(m, " mapped=%lu", md->pages);
1994
1995         if (md->mapcount_max > 1)
1996                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1997
1998         if (md->swapcache)
1999                 seq_printf(m, " swapcache=%lu", md->swapcache);
2000
2001         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2002                 seq_printf(m, " active=%lu", md->active);
2003
2004         if (md->writeback)
2005                 seq_printf(m, " writeback=%lu", md->writeback);
2006
2007         for_each_node_state(nid, N_MEMORY)
2008                 if (md->node[nid])
2009                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2010
2011         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2012 out:
2013         seq_putc(m, '\n');
2014         return 0;
2015 }
2016
2017 static const struct seq_operations proc_pid_numa_maps_op = {
2018         .start  = m_start,
2019         .next   = m_next,
2020         .stop   = m_stop,
2021         .show   = show_numa_map,
2022 };
2023
2024 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2025 {
2026         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2027                                 sizeof(struct numa_maps_private));
2028 }
2029
2030 const struct file_operations proc_pid_numa_maps_operations = {
2031         .open           = pid_numa_maps_open,
2032         .read           = seq_read,
2033         .llseek         = seq_lseek,
2034         .release        = proc_map_release,
2035 };
2036
2037 #endif /* CONFIG_NUMA */