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/seq_file.h>
   8 #include <linux/highmem.h>
   9 #include <linux/ptrace.h>
  10 #include <linux/slab.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/mempolicy.h>
  13 #include <linux/rmap.h>
  14 #include <linux/swap.h>
  15 #include <linux/sched/mm.h>
  16 #include <linux/swapops.h>
  17 #include <linux/mmu_notifier.h>
  18 #include <linux/page_idle.h>
  19 #include <linux/shmem_fs.h>
  20 #include <linux/uaccess.h>
  21 #include <linux/pkeys.h>
  22
  23 #include <asm/elf.h>
  24 #include <asm/tlb.h>
  25 #include <asm/tlbflush.h>
  26 #include "internal.h"
  27
  28 #define SEQ_PUT_DEC(str, val) \
  29                 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
  30 void task_mem(struct seq_file *m, struct mm_struct *mm)
  31 {
  32         unsigned long text, lib, swap, anon, file, shmem;
  33         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  34
  35         anon = get_mm_counter(mm, MM_ANONPAGES);
  36         file = get_mm_counter(mm, MM_FILEPAGES);
  37         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  38
  39         /*
  40          * Note: to minimize their overhead, mm maintains hiwater_vm and
  41          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  42          * collector of these hiwater stats must therefore get total_vm
  43          * and rss too, which will usually be the higher.  Barriers? not
  44          * worth the effort, such snapshots can always be inconsistent.
  45          */
  46         hiwater_vm = total_vm = mm->total_vm;
  47         if (hiwater_vm < mm->hiwater_vm)
  48                 hiwater_vm = mm->hiwater_vm;
  49         hiwater_rss = total_rss = anon + file + shmem;
  50         if (hiwater_rss < mm->hiwater_rss)
  51                 hiwater_rss = mm->hiwater_rss;
  52
  53         /* split executable areas between text and lib */
  54         text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
  55         text = min(text, mm->exec_vm << PAGE_SHIFT);
  56         lib = (mm->exec_vm << PAGE_SHIFT) - text;
  57
  58         swap = get_mm_counter(mm, MM_SWAPENTS);
  59         SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
  60         SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
  61         SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
  62         SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
  63         SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
  64         SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
  65         SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
  66         SEQ_PUT_DEC(" kB\nRssFile:\t", file);
  67         SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
  68         SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
  69         SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
  70         seq_put_decimal_ull_width(m,
  71                     " kB\nVmExe:\t", text >> 10, 8);
  72         seq_put_decimal_ull_width(m,
  73                     " kB\nVmLib:\t", lib >> 10, 8);
  74         seq_put_decimal_ull_width(m,
  75                     " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
  76         SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
  77         seq_puts(m, " kB\n");
  78         hugetlb_report_usage(m, mm);
  79 }
  80 #undef SEQ_PUT_DEC
  81
  82 unsigned long task_vsize(struct mm_struct *mm)
  83 {
  84         return PAGE_SIZE * mm->total_vm;
  85 }
  86
  87 unsigned long task_statm(struct mm_struct *mm,
  88                          unsigned long *shared, unsigned long *text,
  89                          unsigned long *data, unsigned long *resident)
  90 {
  91         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  92                         get_mm_counter(mm, MM_SHMEMPAGES);
  93         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  94                                                                 >> PAGE_SHIFT;
  95         *data = mm->data_vm + mm->stack_vm;
  96         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
  97         return mm->total_vm;
  98 }
  99
 100 #ifdef CONFIG_NUMA
 101 /*
 102  * Save get_task_policy() for show_numa_map().
 103  */
 104 static void hold_task_mempolicy(struct proc_maps_private *priv)
 105 {
 106         struct task_struct *task = priv->task;
 107
 108         task_lock(task);
 109         priv->task_mempolicy = get_task_policy(task);
 110         mpol_get(priv->task_mempolicy);
 111         task_unlock(task);
 112 }
 113 static void release_task_mempolicy(struct proc_maps_private *priv)
 114 {
 115         mpol_put(priv->task_mempolicy);
 116 }
 117 #else
 118 static void hold_task_mempolicy(struct proc_maps_private *priv)
 119 {
 120 }
 121 static void release_task_mempolicy(struct proc_maps_private *priv)
 122 {
 123 }
 124 #endif
 125
 126 static struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv,
 127                                                 loff_t *ppos)
 128 {
 129         struct vm_area_struct *vma = vma_next(&priv->iter);
 130
 131         if (vma) {
 132                 *ppos = vma->vm_start;
 133         } else {
 134                 *ppos = -2UL;
 135                 vma = get_gate_vma(priv->mm);
 136         }
 137
 138         return vma;
 139 }
 140
 141 static void *m_start(struct seq_file *m, loff_t *ppos)
 142 {
 143         struct proc_maps_private *priv = m->private;
 144         unsigned long last_addr = *ppos;
 145         struct mm_struct *mm;
 146
 147         /* See m_next(). Zero at the start or after lseek. */
 148         if (last_addr == -1UL)
 149                 return NULL;
 150
 151         priv->task = get_proc_task(priv->inode);
 152         if (!priv->task)
 153                 return ERR_PTR(-ESRCH);
 154
 155         mm = priv->mm;
 156         if (!mm || !mmget_not_zero(mm)) {
 157                 put_task_struct(priv->task);
 158                 priv->task = NULL;
 159                 return NULL;
 160         }
 161
 162         if (mmap_read_lock_killable(mm)) {
 163                 mmput(mm);
 164                 put_task_struct(priv->task);
 165                 priv->task = NULL;
 166                 return ERR_PTR(-EINTR);
 167         }
 168
 169         vma_iter_init(&priv->iter, mm, last_addr);
 170         hold_task_mempolicy(priv);
 171         if (last_addr == -2UL)
 172                 return get_gate_vma(mm);
 173
 174         return proc_get_vma(priv, ppos);
 175 }
 176
 177 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
 178 {
 179         if (*ppos == -2UL) {
 180                 *ppos = -1UL;
 181                 return NULL;
 182         }
 183         return proc_get_vma(m->private, ppos);
 184 }
 185
 186 static void m_stop(struct seq_file *m, void *v)
 187 {
 188         struct proc_maps_private *priv = m->private;
 189         struct mm_struct *mm = priv->mm;
 190
 191         if (!priv->task)
 192                 return;
 193
 194         release_task_mempolicy(priv);
 195         mmap_read_unlock(mm);
 196         mmput(mm);
 197         put_task_struct(priv->task);
 198         priv->task = NULL;
 199 }
 200
 201 static int proc_maps_open(struct inode *inode, struct file *file,
 202                         const struct seq_operations *ops, int psize)
 203 {
 204         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 205
 206         if (!priv)
 207                 return -ENOMEM;
 208
 209         priv->inode = inode;
 210         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 211         if (IS_ERR(priv->mm)) {
 212                 int err = PTR_ERR(priv->mm);
 213
 214                 seq_release_private(inode, file);
 215                 return err;
 216         }
 217
 218         return 0;
 219 }
 220
 221 static int proc_map_release(struct inode *inode, struct file *file)
 222 {
 223         struct seq_file *seq = file->private_data;
 224         struct proc_maps_private *priv = seq->private;
 225
 226         if (priv->mm)
 227                 mmdrop(priv->mm);
 228
 229         return seq_release_private(inode, file);
 230 }
 231
 232 static int do_maps_open(struct inode *inode, struct file *file,
 233                         const struct seq_operations *ops)
 234 {
 235         return proc_maps_open(inode, file, ops,
 236                                 sizeof(struct proc_maps_private));
 237 }
 238
 239 /*
 240  * Indicate if the VMA is a stack for the given task; for
 241  * /proc/PID/maps that is the stack of the main task.
 242  */
 243 static int is_stack(struct vm_area_struct *vma)
 244 {
 245         /*
 246          * We make no effort to guess what a given thread considers to be
 247          * its "stack".  It's not even well-defined for programs written
 248          * languages like Go.
 249          */
 250         return vma->vm_start <= vma->vm_mm->start_stack &&
 251                 vma->vm_end >= vma->vm_mm->start_stack;
 252 }
 253
 254 static void show_vma_header_prefix(struct seq_file *m,
 255                                    unsigned long start, unsigned long end,
 256                                    vm_flags_t flags, unsigned long long pgoff,
 257                                    dev_t dev, unsigned long ino)
 258 {
 259         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 260         seq_put_hex_ll(m, NULL, start, 8);
 261         seq_put_hex_ll(m, "-", end, 8);
 262         seq_putc(m, ' ');
 263         seq_putc(m, flags & VM_READ ? 'r' : '-');
 264         seq_putc(m, flags & VM_WRITE ? 'w' : '-');
 265         seq_putc(m, flags & VM_EXEC ? 'x' : '-');
 266         seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
 267         seq_put_hex_ll(m, " ", pgoff, 8);
 268         seq_put_hex_ll(m, " ", MAJOR(dev), 2);
 269         seq_put_hex_ll(m, ":", MINOR(dev), 2);
 270         seq_put_decimal_ull(m, " ", ino);
 271         seq_putc(m, ' ');
 272 }
 273
 274 static void
 275 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 276 {
 277         struct anon_vma_name *anon_name = NULL;
 278         struct mm_struct *mm = vma->vm_mm;
 279         struct file *file = vma->vm_file;
 280         vm_flags_t flags = vma->vm_flags;
 281         unsigned long ino = 0;
 282         unsigned long long pgoff = 0;
 283         unsigned long start, end;
 284         dev_t dev = 0;
 285         const char *name = NULL;
 286
 287         if (file) {
 288                 struct inode *inode = file_inode(vma->vm_file);
 289                 dev = inode->i_sb->s_dev;
 290                 ino = inode->i_ino;
 291                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 292         }
 293
 294         start = vma->vm_start;
 295         end = vma->vm_end;
 296         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 297         if (mm)
 298                 anon_name = anon_vma_name(vma);
 299
 300         /*
 301          * Print the dentry name for named mappings, and a
 302          * special [heap] marker for the heap:
 303          */
 304         if (file) {
 305                 seq_pad(m, ' ');
 306                 /*
 307                  * If user named this anon shared memory via
 308                  * prctl(PR_SET_VMA ..., use the provided name.
 309                  */
 310                 if (anon_name)
 311                         seq_printf(m, "[anon_shmem:%s]", anon_name->name);
 312                 else
 313                         seq_file_path(m, file, "\n");
 314                 goto done;
 315         }
 316
 317         if (vma->vm_ops && vma->vm_ops->name) {
 318                 name = vma->vm_ops->name(vma);
 319                 if (name)
 320                         goto done;
 321         }
 322
 323         name = arch_vma_name(vma);
 324         if (!name) {
 325                 if (!mm) {
 326                         name = "[vdso]";
 327                         goto done;
 328                 }
 329
 330                 if (vma->vm_start <= mm->brk &&
 331                     vma->vm_end >= mm->start_brk) {
 332                         name = "[heap]";
 333                         goto done;
 334                 }
 335
 336                 if (is_stack(vma)) {
 337                         name = "[stack]";
 338                         goto done;
 339                 }
 340
 341                 if (anon_name) {
 342                         seq_pad(m, ' ');
 343                         seq_printf(m, "[anon:%s]", anon_name->name);
 344                 }
 345         }
 346
 347 done:
 348         if (name) {
 349                 seq_pad(m, ' ');
 350                 seq_puts(m, name);
 351         }
 352         seq_putc(m, '\n');
 353 }
 354
 355 static int show_map(struct seq_file *m, void *v)
 356 {
 357         show_map_vma(m, v);
 358         return 0;
 359 }
 360
 361 static const struct seq_operations proc_pid_maps_op = {
 362         .start  = m_start,
 363         .next   = m_next,
 364         .stop   = m_stop,
 365         .show   = show_map
 366 };
 367
 368 static int pid_maps_open(struct inode *inode, struct file *file)
 369 {
 370         return do_maps_open(inode, file, &proc_pid_maps_op);
 371 }
 372
 373 const struct file_operations proc_pid_maps_operations = {
 374         .open           = pid_maps_open,
 375         .read           = seq_read,
 376         .llseek         = seq_lseek,
 377         .release        = proc_map_release,
 378 };
 379
 380 /*
 381  * Proportional Set Size(PSS): my share of RSS.
 382  *
 383  * PSS of a process is the count of pages it has in memory, where each
 384  * page is divided by the number of processes sharing it.  So if a
 385  * process has 1000 pages all to itself, and 1000 shared with one other
 386  * process, its PSS will be 1500.
 387  *
 388  * To keep (accumulated) division errors low, we adopt a 64bit
 389  * fixed-point pss counter to minimize division errors. So (pss >>
 390  * PSS_SHIFT) would be the real byte count.
 391  *
 392  * A shift of 12 before division means (assuming 4K page size):
 393  *      - 1M 3-user-pages add up to 8KB errors;
 394  *      - supports mapcount up to 2^24, or 16M;
 395  *      - supports PSS up to 2^52 bytes, or 4PB.
 396  */
 397 #define PSS_SHIFT 12
 398
 399 #ifdef CONFIG_PROC_PAGE_MONITOR
 400 struct mem_size_stats {
 401         unsigned long resident;
 402         unsigned long shared_clean;
 403         unsigned long shared_dirty;
 404         unsigned long private_clean;
 405         unsigned long private_dirty;
 406         unsigned long referenced;
 407         unsigned long anonymous;
 408         unsigned long lazyfree;
 409         unsigned long anonymous_thp;
 410         unsigned long shmem_thp;
 411         unsigned long file_thp;
 412         unsigned long swap;
 413         unsigned long shared_hugetlb;
 414         unsigned long private_hugetlb;
 415         u64 pss;
 416         u64 pss_anon;
 417         u64 pss_file;
 418         u64 pss_shmem;
 419         u64 pss_dirty;
 420         u64 pss_locked;
 421         u64 swap_pss;
 422 };
 423
 424 static void smaps_page_accumulate(struct mem_size_stats *mss,
 425                 struct page *page, unsigned long size, unsigned long pss,
 426                 bool dirty, bool locked, bool private)
 427 {
 428         mss->pss += pss;
 429
 430         if (PageAnon(page))
 431                 mss->pss_anon += pss;
 432         else if (PageSwapBacked(page))
 433                 mss->pss_shmem += pss;
 434         else
 435                 mss->pss_file += pss;
 436
 437         if (locked)
 438                 mss->pss_locked += pss;
 439
 440         if (dirty || PageDirty(page)) {
 441                 mss->pss_dirty += pss;
 442                 if (private)
 443                         mss->private_dirty += size;
 444                 else
 445                         mss->shared_dirty += size;
 446         } else {
 447                 if (private)
 448                         mss->private_clean += size;
 449                 else
 450                         mss->shared_clean += size;
 451         }
 452 }
 453
 454 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 455                 bool compound, bool young, bool dirty, bool locked,
 456                 bool migration)
 457 {
 458         int i, nr = compound ? compound_nr(page) : 1;
 459         unsigned long size = nr * PAGE_SIZE;
 460
 461         /*
 462          * First accumulate quantities that depend only on |size| and the type
 463          * of the compound page.
 464          */
 465         if (PageAnon(page)) {
 466                 mss->anonymous += size;
 467                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 468                         mss->lazyfree += size;
 469         }
 470
 471         mss->resident += size;
 472         /* Accumulate the size in pages that have been accessed. */
 473         if (young || page_is_young(page) || PageReferenced(page))
 474                 mss->referenced += size;
 475
 476         /*
 477          * Then accumulate quantities that may depend on sharing, or that may
 478          * differ page-by-page.
 479          *
 480          * page_count(page) == 1 guarantees the page is mapped exactly once.
 481          * If any subpage of the compound page mapped with PTE it would elevate
 482          * page_count().
 483          *
 484          * The page_mapcount() is called to get a snapshot of the mapcount.
 485          * Without holding the page lock this snapshot can be slightly wrong as
 486          * we cannot always read the mapcount atomically.  It is not safe to
 487          * call page_mapcount() even with PTL held if the page is not mapped,
 488          * especially for migration entries.  Treat regular migration entries
 489          * as mapcount == 1.
 490          */
 491         if ((page_count(page) == 1) || migration) {
 492                 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
 493                         locked, true);
 494                 return;
 495         }
 496         for (i = 0; i < nr; i++, page++) {
 497                 int mapcount = page_mapcount(page);
 498                 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
 499                 if (mapcount >= 2)
 500                         pss /= mapcount;
 501                 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
 502                                       mapcount < 2);
 503         }
 504 }
 505
 506 #ifdef CONFIG_SHMEM
 507 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 508                           __always_unused int depth, struct mm_walk *walk)
 509 {
 510         struct mem_size_stats *mss = walk->private;
 511         struct vm_area_struct *vma = walk->vma;
 512
 513         mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping,
 514                                               linear_page_index(vma, addr),
 515                                               linear_page_index(vma, end));
 516
 517         return 0;
 518 }
 519 #else
 520 #define smaps_pte_hole          NULL
 521 #endif /* CONFIG_SHMEM */
 522
 523 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
 524 {
 525 #ifdef CONFIG_SHMEM
 526         if (walk->ops->pte_hole) {
 527                 /* depth is not used */
 528                 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
 529         }
 530 #endif
 531 }
 532
 533 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 534                 struct mm_walk *walk)
 535 {
 536         struct mem_size_stats *mss = walk->private;
 537         struct vm_area_struct *vma = walk->vma;
 538         bool locked = !!(vma->vm_flags & VM_LOCKED);
 539         struct page *page = NULL;
 540         bool migration = false, young = false, dirty = false;
 541
 542         if (pte_present(*pte)) {
 543                 page = vm_normal_page(vma, addr, *pte);
 544                 young = pte_young(*pte);
 545                 dirty = pte_dirty(*pte);
 546         } else if (is_swap_pte(*pte)) {
 547                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 548
 549                 if (!non_swap_entry(swpent)) {
 550                         int mapcount;
 551
 552                         mss->swap += PAGE_SIZE;
 553                         mapcount = swp_swapcount(swpent);
 554                         if (mapcount >= 2) {
 555                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 556
 557                                 do_div(pss_delta, mapcount);
 558                                 mss->swap_pss += pss_delta;
 559                         } else {
 560                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 561                         }
 562                 } else if (is_pfn_swap_entry(swpent)) {
 563                         if (is_migration_entry(swpent))
 564                                 migration = true;
 565                         page = pfn_swap_entry_to_page(swpent);
 566                 }
 567         } else {
 568                 smaps_pte_hole_lookup(addr, walk);
 569                 return;
 570         }
 571
 572         if (!page)
 573                 return;
 574
 575         smaps_account(mss, page, false, young, dirty, locked, migration);
 576 }
 577
 578 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 579 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 580                 struct mm_walk *walk)
 581 {
 582         struct mem_size_stats *mss = walk->private;
 583         struct vm_area_struct *vma = walk->vma;
 584         bool locked = !!(vma->vm_flags & VM_LOCKED);
 585         struct page *page = NULL;
 586         bool migration = false;
 587
 588         if (pmd_present(*pmd)) {
 589                 /* FOLL_DUMP will return -EFAULT on huge zero page */
 590                 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 591         } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
 592                 swp_entry_t entry = pmd_to_swp_entry(*pmd);
 593
 594                 if (is_migration_entry(entry)) {
 595                         migration = true;
 596                         page = pfn_swap_entry_to_page(entry);
 597                 }
 598         }
 599         if (IS_ERR_OR_NULL(page))
 600                 return;
 601         if (PageAnon(page))
 602                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 603         else if (PageSwapBacked(page))
 604                 mss->shmem_thp += HPAGE_PMD_SIZE;
 605         else if (is_zone_device_page(page))
 606                 /* pass */;
 607         else
 608                 mss->file_thp += HPAGE_PMD_SIZE;
 609
 610         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
 611                       locked, migration);
 612 }
 613 #else
 614 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 615                 struct mm_walk *walk)
 616 {
 617 }
 618 #endif
 619
 620 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 621                            struct mm_walk *walk)
 622 {
 623         struct vm_area_struct *vma = walk->vma;
 624         pte_t *pte;
 625         spinlock_t *ptl;
 626
 627         ptl = pmd_trans_huge_lock(pmd, vma);
 628         if (ptl) {
 629                 smaps_pmd_entry(pmd, addr, walk);
 630                 spin_unlock(ptl);
 631                 goto out;
 632         }
 633
 634         if (pmd_trans_unstable(pmd))
 635                 goto out;
 636         /*
 637          * The mmap_lock held all the way back in m_start() is what
 638          * keeps khugepaged out of here and from collapsing things
 639          * in here.
 640          */
 641         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 642         for (; addr != end; pte++, addr += PAGE_SIZE)
 643                 smaps_pte_entry(pte, addr, walk);
 644         pte_unmap_unlock(pte - 1, ptl);
 645 out:
 646         cond_resched();
 647         return 0;
 648 }
 649
 650 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 651 {
 652         /*
 653          * Don't forget to update Documentation/ on changes.
 654          */
 655         static const char mnemonics[BITS_PER_LONG][2] = {
 656                 /*
 657                  * In case if we meet a flag we don't know about.
 658                  */
 659                 [0 ... (BITS_PER_LONG-1)] = "??",
 660
 661                 [ilog2(VM_READ)]        = "rd",
 662                 [ilog2(VM_WRITE)]       = "wr",
 663                 [ilog2(VM_EXEC)]        = "ex",
 664                 [ilog2(VM_SHARED)]      = "sh",
 665                 [ilog2(VM_MAYREAD)]     = "mr",
 666                 [ilog2(VM_MAYWRITE)]    = "mw",
 667                 [ilog2(VM_MAYEXEC)]     = "me",
 668                 [ilog2(VM_MAYSHARE)]    = "ms",
 669                 [ilog2(VM_GROWSDOWN)]   = "gd",
 670                 [ilog2(VM_PFNMAP)]      = "pf",
 671                 [ilog2(VM_LOCKED)]      = "lo",
 672                 [ilog2(VM_IO)]          = "io",
 673                 [ilog2(VM_SEQ_READ)]    = "sr",
 674                 [ilog2(VM_RAND_READ)]   = "rr",
 675                 [ilog2(VM_DONTCOPY)]    = "dc",
 676                 [ilog2(VM_DONTEXPAND)]  = "de",
 677                 [ilog2(VM_LOCKONFAULT)] = "lf",
 678                 [ilog2(VM_ACCOUNT)]     = "ac",
 679                 [ilog2(VM_NORESERVE)]   = "nr",
 680                 [ilog2(VM_HUGETLB)]     = "ht",
 681                 [ilog2(VM_SYNC)]        = "sf",
 682                 [ilog2(VM_ARCH_1)]      = "ar",
 683                 [ilog2(VM_WIPEONFORK)]  = "wf",
 684                 [ilog2(VM_DONTDUMP)]    = "dd",
 685 #ifdef CONFIG_ARM64_BTI
 686                 [ilog2(VM_ARM64_BTI)]   = "bt",
 687 #endif
 688 #ifdef CONFIG_MEM_SOFT_DIRTY
 689                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 690 #endif
 691                 [ilog2(VM_MIXEDMAP)]    = "mm",
 692                 [ilog2(VM_HUGEPAGE)]    = "hg",
 693                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 694                 [ilog2(VM_MERGEABLE)]   = "mg",
 695                 [ilog2(VM_UFFD_MISSING)]= "um",
 696                 [ilog2(VM_UFFD_WP)]     = "uw",
 697 #ifdef CONFIG_ARM64_MTE
 698                 [ilog2(VM_MTE)]         = "mt",
 699                 [ilog2(VM_MTE_ALLOWED)] = "",
 700 #endif
 701 #ifdef CONFIG_ARCH_HAS_PKEYS
 702                 /* These come out via ProtectionKey: */
 703                 [ilog2(VM_PKEY_BIT0)]   = "",
 704                 [ilog2(VM_PKEY_BIT1)]   = "",
 705                 [ilog2(VM_PKEY_BIT2)]   = "",
 706                 [ilog2(VM_PKEY_BIT3)]   = "",
 707 #if VM_PKEY_BIT4
 708                 [ilog2(VM_PKEY_BIT4)]   = "",
 709 #endif
 710 #endif /* CONFIG_ARCH_HAS_PKEYS */
 711 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
 712                 [ilog2(VM_UFFD_MINOR)]  = "ui",
 713 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
 714         };
 715         size_t i;
 716
 717         seq_puts(m, "VmFlags: ");
 718         for (i = 0; i < BITS_PER_LONG; i++) {
 719                 if (!mnemonics[i][0])
 720                         continue;
 721                 if (vma->vm_flags & (1UL << i)) {
 722                         seq_putc(m, mnemonics[i][0]);
 723                         seq_putc(m, mnemonics[i][1]);
 724                         seq_putc(m, ' ');
 725                 }
 726         }
 727         seq_putc(m, '\n');
 728 }
 729
 730 #ifdef CONFIG_HUGETLB_PAGE
 731 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 732                                  unsigned long addr, unsigned long end,
 733                                  struct mm_walk *walk)
 734 {
 735         struct mem_size_stats *mss = walk->private;
 736         struct vm_area_struct *vma = walk->vma;
 737         struct page *page = NULL;
 738
 739         if (pte_present(*pte)) {
 740                 page = vm_normal_page(vma, addr, *pte);
 741         } else if (is_swap_pte(*pte)) {
 742                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 743
 744                 if (is_pfn_swap_entry(swpent))
 745                         page = pfn_swap_entry_to_page(swpent);
 746         }
 747         if (page) {
 748                 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
 749                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 750                 else
 751                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 752         }
 753         return 0;
 754 }
 755 #else
 756 #define smaps_hugetlb_range     NULL
 757 #endif /* HUGETLB_PAGE */
 758
 759 static const struct mm_walk_ops smaps_walk_ops = {
 760         .pmd_entry              = smaps_pte_range,
 761         .hugetlb_entry          = smaps_hugetlb_range,
 762 };
 763
 764 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 765         .pmd_entry              = smaps_pte_range,
 766         .hugetlb_entry          = smaps_hugetlb_range,
 767         .pte_hole               = smaps_pte_hole,
 768 };
 769
 770 /*
 771  * Gather mem stats from @vma with the indicated beginning
 772  * address @start, and keep them in @mss.
 773  *
 774  * Use vm_start of @vma as the beginning address if @start is 0.
 775  */
 776 static void smap_gather_stats(struct vm_area_struct *vma,
 777                 struct mem_size_stats *mss, unsigned long start)
 778 {
 779         const struct mm_walk_ops *ops = &smaps_walk_ops;
 780
 781         /* Invalid start */
 782         if (start >= vma->vm_end)
 783                 return;
 784
 785 #ifdef CONFIG_SHMEM
 786         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 787                 /*
 788                  * For shared or readonly shmem mappings we know that all
 789                  * swapped out pages belong to the shmem object, and we can
 790                  * obtain the swap value much more efficiently. For private
 791                  * writable mappings, we might have COW pages that are
 792                  * not affected by the parent swapped out pages of the shmem
 793                  * object, so we have to distinguish them during the page walk.
 794                  * Unless we know that the shmem object (or the part mapped by
 795                  * our VMA) has no swapped out pages at all.
 796                  */
 797                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 798
 799                 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 800                                         !(vma->vm_flags & VM_WRITE))) {
 801                         mss->swap += shmem_swapped;
 802                 } else {
 803                         ops = &smaps_shmem_walk_ops;
 804                 }
 805         }
 806 #endif
 807         /* mmap_lock is held in m_start */
 808         if (!start)
 809                 walk_page_vma(vma, ops, mss);
 810         else
 811                 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
 812 }
 813
 814 #define SEQ_PUT_DEC(str, val) \
 815                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 816
 817 /* Show the contents common for smaps and smaps_rollup */
 818 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 819         bool rollup_mode)
 820 {
 821         SEQ_PUT_DEC("Rss:            ", mss->resident);
 822         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 823         SEQ_PUT_DEC(" kB\nPss_Dirty:      ", mss->pss_dirty >> PSS_SHIFT);
 824         if (rollup_mode) {
 825                 /*
 826                  * These are meaningful only for smaps_rollup, otherwise two of
 827                  * them are zero, and the other one is the same as Pss.
 828                  */
 829                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 830                         mss->pss_anon >> PSS_SHIFT);
 831                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 832                         mss->pss_file >> PSS_SHIFT);
 833                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 834                         mss->pss_shmem >> PSS_SHIFT);
 835         }
 836         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 837         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 838         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 839         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 840         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 841         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 842         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 843         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 844         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 845         SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
 846         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 847         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 848                                   mss->private_hugetlb >> 10, 7);
 849         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 850         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 851                                         mss->swap_pss >> PSS_SHIFT);
 852         SEQ_PUT_DEC(" kB\nLocked:         ",
 853                                         mss->pss_locked >> PSS_SHIFT);
 854         seq_puts(m, " kB\n");
 855 }
 856
 857 static int show_smap(struct seq_file *m, void *v)
 858 {
 859         struct vm_area_struct *vma = v;
 860         struct mem_size_stats mss;
 861
 862         memset(&mss, 0, sizeof(mss));
 863
 864         smap_gather_stats(vma, &mss, 0);
 865
 866         show_map_vma(m, vma);
 867
 868         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 869         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 870         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 871         seq_puts(m, " kB\n");
 872
 873         __show_smap(m, &mss, false);
 874
 875         seq_printf(m, "THPeligible:    %d\n",
 876                    hugepage_vma_check(vma, vma->vm_flags, true, false, true));
 877
 878         if (arch_pkeys_enabled())
 879                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 880         show_smap_vma_flags(m, vma);
 881
 882         return 0;
 883 }
 884
 885 static int show_smaps_rollup(struct seq_file *m, void *v)
 886 {
 887         struct proc_maps_private *priv = m->private;
 888         struct mem_size_stats mss;
 889         struct mm_struct *mm = priv->mm;
 890         struct vm_area_struct *vma;
 891         unsigned long vma_start = 0, last_vma_end = 0;
 892         int ret = 0;
 893         MA_STATE(mas, &mm->mm_mt, 0, 0);
 894
 895         priv->task = get_proc_task(priv->inode);
 896         if (!priv->task)
 897                 return -ESRCH;
 898
 899         if (!mm || !mmget_not_zero(mm)) {
 900                 ret = -ESRCH;
 901                 goto out_put_task;
 902         }
 903
 904         memset(&mss, 0, sizeof(mss));
 905
 906         ret = mmap_read_lock_killable(mm);
 907         if (ret)
 908                 goto out_put_mm;
 909
 910         hold_task_mempolicy(priv);
 911         vma = mas_find(&mas, ULONG_MAX);
 912
 913         if (unlikely(!vma))
 914                 goto empty_set;
 915
 916         vma_start = vma->vm_start;
 917         do {
 918                 smap_gather_stats(vma, &mss, 0);
 919                 last_vma_end = vma->vm_end;
 920
 921                 /*
 922                  * Release mmap_lock temporarily if someone wants to
 923                  * access it for write request.
 924                  */
 925                 if (mmap_lock_is_contended(mm)) {
 926                         mas_pause(&mas);
 927                         mmap_read_unlock(mm);
 928                         ret = mmap_read_lock_killable(mm);
 929                         if (ret) {
 930                                 release_task_mempolicy(priv);
 931                                 goto out_put_mm;
 932                         }
 933
 934                         /*
 935                          * After dropping the lock, there are four cases to
 936                          * consider. See the following example for explanation.
 937                          *
 938                          *   +------+------+-----------+
 939                          *   | VMA1 | VMA2 | VMA3      |
 940                          *   +------+------+-----------+
 941                          *   |      |      |           |
 942                          *  4k     8k     16k         400k
 943                          *
 944                          * Suppose we drop the lock after reading VMA2 due to
 945                          * contention, then we get:
 946                          *
 947                          *      last_vma_end = 16k
 948                          *
 949                          * 1) VMA2 is freed, but VMA3 exists:
 950                          *
 951                          *    find_vma(mm, 16k - 1) will return VMA3.
 952                          *    In this case, just continue from VMA3.
 953                          *
 954                          * 2) VMA2 still exists:
 955                          *
 956                          *    find_vma(mm, 16k - 1) will return VMA2.
 957                          *    Iterate the loop like the original one.
 958                          *
 959                          * 3) No more VMAs can be found:
 960                          *
 961                          *    find_vma(mm, 16k - 1) will return NULL.
 962                          *    No more things to do, just break.
 963                          *
 964                          * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
 965                          *
 966                          *    find_vma(mm, 16k - 1) will return VMA' whose range
 967                          *    contains last_vma_end.
 968                          *    Iterate VMA' from last_vma_end.
 969                          */
 970                         vma = mas_find(&mas, ULONG_MAX);
 971                         /* Case 3 above */
 972                         if (!vma)
 973                                 break;
 974
 975                         /* Case 1 above */
 976                         if (vma->vm_start >= last_vma_end)
 977                                 continue;
 978
 979                         /* Case 4 above */
 980                         if (vma->vm_end > last_vma_end)
 981                                 smap_gather_stats(vma, &mss, last_vma_end);
 982                 }
 983                 /* Case 2 above */
 984         } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL);
 985
 986 empty_set:
 987         show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
 988         seq_pad(m, ' ');
 989         seq_puts(m, "[rollup]\n");
 990
 991         __show_smap(m, &mss, true);
 992
 993         release_task_mempolicy(priv);
 994         mmap_read_unlock(mm);
 995
 996 out_put_mm:
 997         mmput(mm);
 998 out_put_task:
 999         put_task_struct(priv->task);
1000         priv->task = NULL;
1001
1002         return ret;
1003 }
1004 #undef SEQ_PUT_DEC
1005
1006 static const struct seq_operations proc_pid_smaps_op = {
1007         .start  = m_start,
1008         .next   = m_next,
1009         .stop   = m_stop,
1010         .show   = show_smap
1011 };
1012
1013 static int pid_smaps_open(struct inode *inode, struct file *file)
1014 {
1015         return do_maps_open(inode, file, &proc_pid_smaps_op);
1016 }
1017
1018 static int smaps_rollup_open(struct inode *inode, struct file *file)
1019 {
1020         int ret;
1021         struct proc_maps_private *priv;
1022
1023         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1024         if (!priv)
1025                 return -ENOMEM;
1026
1027         ret = single_open(file, show_smaps_rollup, priv);
1028         if (ret)
1029                 goto out_free;
1030
1031         priv->inode = inode;
1032         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1033         if (IS_ERR(priv->mm)) {
1034                 ret = PTR_ERR(priv->mm);
1035
1036                 single_release(inode, file);
1037                 goto out_free;
1038         }
1039
1040         return 0;
1041
1042 out_free:
1043         kfree(priv);
1044         return ret;
1045 }
1046
1047 static int smaps_rollup_release(struct inode *inode, struct file *file)
1048 {
1049         struct seq_file *seq = file->private_data;
1050         struct proc_maps_private *priv = seq->private;
1051
1052         if (priv->mm)
1053                 mmdrop(priv->mm);
1054
1055         kfree(priv);
1056         return single_release(inode, file);
1057 }
1058
1059 const struct file_operations proc_pid_smaps_operations = {
1060         .open           = pid_smaps_open,
1061         .read           = seq_read,
1062         .llseek         = seq_lseek,
1063         .release        = proc_map_release,
1064 };
1065
1066 const struct file_operations proc_pid_smaps_rollup_operations = {
1067         .open           = smaps_rollup_open,
1068         .read           = seq_read,
1069         .llseek         = seq_lseek,
1070         .release        = smaps_rollup_release,
1071 };
1072
1073 enum clear_refs_types {
1074         CLEAR_REFS_ALL = 1,
1075         CLEAR_REFS_ANON,
1076         CLEAR_REFS_MAPPED,
1077         CLEAR_REFS_SOFT_DIRTY,
1078         CLEAR_REFS_MM_HIWATER_RSS,
1079         CLEAR_REFS_LAST,
1080 };
1081
1082 struct clear_refs_private {
1083         enum clear_refs_types type;
1084 };
1085
1086 #ifdef CONFIG_MEM_SOFT_DIRTY
1087
1088 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1089 {
1090         struct page *page;
1091
1092         if (!pte_write(pte))
1093                 return false;
1094         if (!is_cow_mapping(vma->vm_flags))
1095                 return false;
1096         if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1097                 return false;
1098         page = vm_normal_page(vma, addr, pte);
1099         if (!page)
1100                 return false;
1101         return page_maybe_dma_pinned(page);
1102 }
1103
1104 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1105                 unsigned long addr, pte_t *pte)
1106 {
1107         /*
1108          * The soft-dirty tracker uses #PF-s to catch writes
1109          * to pages, so write-protect the pte as well. See the
1110          * Documentation/admin-guide/mm/soft-dirty.rst for full description
1111          * of how soft-dirty works.
1112          */
1113         pte_t ptent = *pte;
1114
1115         if (pte_present(ptent)) {
1116                 pte_t old_pte;
1117
1118                 if (pte_is_pinned(vma, addr, ptent))
1119                         return;
1120                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1121                 ptent = pte_wrprotect(old_pte);
1122                 ptent = pte_clear_soft_dirty(ptent);
1123                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1124         } else if (is_swap_pte(ptent)) {
1125                 ptent = pte_swp_clear_soft_dirty(ptent);
1126                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1127         }
1128 }
1129 #else
1130 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1131                 unsigned long addr, pte_t *pte)
1132 {
1133 }
1134 #endif
1135
1136 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1137 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1138                 unsigned long addr, pmd_t *pmdp)
1139 {
1140         pmd_t old, pmd = *pmdp;
1141
1142         if (pmd_present(pmd)) {
1143                 /* See comment in change_huge_pmd() */
1144                 old = pmdp_invalidate(vma, addr, pmdp);
1145                 if (pmd_dirty(old))
1146                         pmd = pmd_mkdirty(pmd);
1147                 if (pmd_young(old))
1148                         pmd = pmd_mkyoung(pmd);
1149
1150                 pmd = pmd_wrprotect(pmd);
1151                 pmd = pmd_clear_soft_dirty(pmd);
1152
1153                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1154         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1155                 pmd = pmd_swp_clear_soft_dirty(pmd);
1156                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1157         }
1158 }
1159 #else
1160 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1161                 unsigned long addr, pmd_t *pmdp)
1162 {
1163 }
1164 #endif
1165
1166 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1167                                 unsigned long end, struct mm_walk *walk)
1168 {
1169         struct clear_refs_private *cp = walk->private;
1170         struct vm_area_struct *vma = walk->vma;
1171         pte_t *pte, ptent;
1172         spinlock_t *ptl;
1173         struct page *page;
1174
1175         ptl = pmd_trans_huge_lock(pmd, vma);
1176         if (ptl) {
1177                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1178                         clear_soft_dirty_pmd(vma, addr, pmd);
1179                         goto out;
1180                 }
1181
1182                 if (!pmd_present(*pmd))
1183                         goto out;
1184
1185                 page = pmd_page(*pmd);
1186
1187                 /* Clear accessed and referenced bits. */
1188                 pmdp_test_and_clear_young(vma, addr, pmd);
1189                 test_and_clear_page_young(page);
1190                 ClearPageReferenced(page);
1191 out:
1192                 spin_unlock(ptl);
1193                 return 0;
1194         }
1195
1196         if (pmd_trans_unstable(pmd))
1197                 return 0;
1198
1199         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1200         for (; addr != end; pte++, addr += PAGE_SIZE) {
1201                 ptent = *pte;
1202
1203                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1204                         clear_soft_dirty(vma, addr, pte);
1205                         continue;
1206                 }
1207
1208                 if (!pte_present(ptent))
1209                         continue;
1210
1211                 page = vm_normal_page(vma, addr, ptent);
1212                 if (!page)
1213                         continue;
1214
1215                 /* Clear accessed and referenced bits. */
1216                 ptep_test_and_clear_young(vma, addr, pte);
1217                 test_and_clear_page_young(page);
1218                 ClearPageReferenced(page);
1219         }
1220         pte_unmap_unlock(pte - 1, ptl);
1221         cond_resched();
1222         return 0;
1223 }
1224
1225 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1226                                 struct mm_walk *walk)
1227 {
1228         struct clear_refs_private *cp = walk->private;
1229         struct vm_area_struct *vma = walk->vma;
1230
1231         if (vma->vm_flags & VM_PFNMAP)
1232                 return 1;
1233
1234         /*
1235          * Writing 1 to /proc/pid/clear_refs affects all pages.
1236          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1237          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1238          * Writing 4 to /proc/pid/clear_refs affects all pages.
1239          */
1240         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1241                 return 1;
1242         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1243                 return 1;
1244         return 0;
1245 }
1246
1247 static const struct mm_walk_ops clear_refs_walk_ops = {
1248         .pmd_entry              = clear_refs_pte_range,
1249         .test_walk              = clear_refs_test_walk,
1250 };
1251
1252 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1253                                 size_t count, loff_t *ppos)
1254 {
1255         struct task_struct *task;
1256         char buffer[PROC_NUMBUF];
1257         struct mm_struct *mm;
1258         struct vm_area_struct *vma;
1259         enum clear_refs_types type;
1260         int itype;
1261         int rv;
1262
1263         memset(buffer, 0, sizeof(buffer));
1264         if (count > sizeof(buffer) - 1)
1265                 count = sizeof(buffer) - 1;
1266         if (copy_from_user(buffer, buf, count))
1267                 return -EFAULT;
1268         rv = kstrtoint(strstrip(buffer), 10, &itype);
1269         if (rv < 0)
1270                 return rv;
1271         type = (enum clear_refs_types)itype;
1272         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1273                 return -EINVAL;
1274
1275         task = get_proc_task(file_inode(file));
1276         if (!task)
1277                 return -ESRCH;
1278         mm = get_task_mm(task);
1279         if (mm) {
1280                 MA_STATE(mas, &mm->mm_mt, 0, 0);
1281                 struct mmu_notifier_range range;
1282                 struct clear_refs_private cp = {
1283                         .type = type,
1284                 };
1285
1286                 if (mmap_write_lock_killable(mm)) {
1287                         count = -EINTR;
1288                         goto out_mm;
1289                 }
1290                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1291                         /*
1292                          * Writing 5 to /proc/pid/clear_refs resets the peak
1293                          * resident set size to this mm's current rss value.
1294                          */
1295                         reset_mm_hiwater_rss(mm);
1296                         goto out_unlock;
1297                 }
1298
1299                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1300                         mas_for_each(&mas, vma, ULONG_MAX) {
1301                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1302                                         continue;
1303                                 vma->vm_flags &= ~VM_SOFTDIRTY;
1304                                 vma_set_page_prot(vma);
1305                         }
1306
1307                         inc_tlb_flush_pending(mm);
1308                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1309                                                 0, NULL, mm, 0, -1UL);
1310                         mmu_notifier_invalidate_range_start(&range);
1311                 }
1312                 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1313                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1314                         mmu_notifier_invalidate_range_end(&range);
1315                         flush_tlb_mm(mm);
1316                         dec_tlb_flush_pending(mm);
1317                 }
1318 out_unlock:
1319                 mmap_write_unlock(mm);
1320 out_mm:
1321                 mmput(mm);
1322         }
1323         put_task_struct(task);
1324
1325         return count;
1326 }
1327
1328 const struct file_operations proc_clear_refs_operations = {
1329         .write          = clear_refs_write,
1330         .llseek         = noop_llseek,
1331 };
1332
1333 typedef struct {
1334         u64 pme;
1335 } pagemap_entry_t;
1336
1337 struct pagemapread {
1338         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1339         pagemap_entry_t *buffer;
1340         bool show_pfn;
1341 };
1342
1343 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1344 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1345
1346 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1347 #define PM_PFRAME_BITS          55
1348 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1349 #define PM_SOFT_DIRTY           BIT_ULL(55)
1350 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1351 #define PM_UFFD_WP              BIT_ULL(57)
1352 #define PM_FILE                 BIT_ULL(61)
1353 #define PM_SWAP                 BIT_ULL(62)
1354 #define PM_PRESENT              BIT_ULL(63)
1355
1356 #define PM_END_OF_BUFFER    1
1357
1358 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1359 {
1360         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1361 }
1362
1363 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1364                           struct pagemapread *pm)
1365 {
1366         pm->buffer[pm->pos++] = *pme;
1367         if (pm->pos >= pm->len)
1368                 return PM_END_OF_BUFFER;
1369         return 0;
1370 }
1371
1372 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1373                             __always_unused int depth, struct mm_walk *walk)
1374 {
1375         struct pagemapread *pm = walk->private;
1376         unsigned long addr = start;
1377         int err = 0;
1378
1379         while (addr < end) {
1380                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1381                 pagemap_entry_t pme = make_pme(0, 0);
1382                 /* End of address space hole, which we mark as non-present. */
1383                 unsigned long hole_end;
1384
1385                 if (vma)
1386                         hole_end = min(end, vma->vm_start);
1387                 else
1388                         hole_end = end;
1389
1390                 for (; addr < hole_end; addr += PAGE_SIZE) {
1391                         err = add_to_pagemap(addr, &pme, pm);
1392                         if (err)
1393                                 goto out;
1394                 }
1395
1396                 if (!vma)
1397                         break;
1398
1399                 /* Addresses in the VMA. */
1400                 if (vma->vm_flags & VM_SOFTDIRTY)
1401                         pme = make_pme(0, PM_SOFT_DIRTY);
1402                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1403                         err = add_to_pagemap(addr, &pme, pm);
1404                         if (err)
1405                                 goto out;
1406                 }
1407         }
1408 out:
1409         return err;
1410 }
1411
1412 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1413                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1414 {
1415         u64 frame = 0, flags = 0;
1416         struct page *page = NULL;
1417         bool migration = false;
1418
1419         if (pte_present(pte)) {
1420                 if (pm->show_pfn)
1421                         frame = pte_pfn(pte);
1422                 flags |= PM_PRESENT;
1423                 page = vm_normal_page(vma, addr, pte);
1424                 if (pte_soft_dirty(pte))
1425                         flags |= PM_SOFT_DIRTY;
1426                 if (pte_uffd_wp(pte))
1427                         flags |= PM_UFFD_WP;
1428         } else if (is_swap_pte(pte)) {
1429                 swp_entry_t entry;
1430                 if (pte_swp_soft_dirty(pte))
1431                         flags |= PM_SOFT_DIRTY;
1432                 if (pte_swp_uffd_wp(pte))
1433                         flags |= PM_UFFD_WP;
1434                 entry = pte_to_swp_entry(pte);
1435                 if (pm->show_pfn) {
1436                         pgoff_t offset;
1437                         /*
1438                          * For PFN swap offsets, keeping the offset field
1439                          * to be PFN only to be compatible with old smaps.
1440                          */
1441                         if (is_pfn_swap_entry(entry))
1442                                 offset = swp_offset_pfn(entry);
1443                         else
1444                                 offset = swp_offset(entry);
1445                         frame = swp_type(entry) |
1446                             (offset << MAX_SWAPFILES_SHIFT);
1447                 }
1448                 flags |= PM_SWAP;
1449                 migration = is_migration_entry(entry);
1450                 if (is_pfn_swap_entry(entry))
1451                         page = pfn_swap_entry_to_page(entry);
1452                 if (pte_marker_entry_uffd_wp(entry))
1453                         flags |= PM_UFFD_WP;
1454         }
1455
1456         if (page && !PageAnon(page))
1457                 flags |= PM_FILE;
1458         if (page && !migration && page_mapcount(page) == 1)
1459                 flags |= PM_MMAP_EXCLUSIVE;
1460         if (vma->vm_flags & VM_SOFTDIRTY)
1461                 flags |= PM_SOFT_DIRTY;
1462
1463         return make_pme(frame, flags);
1464 }
1465
1466 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1467                              struct mm_walk *walk)
1468 {
1469         struct vm_area_struct *vma = walk->vma;
1470         struct pagemapread *pm = walk->private;
1471         spinlock_t *ptl;
1472         pte_t *pte, *orig_pte;
1473         int err = 0;
1474 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1475         bool migration = false;
1476
1477         ptl = pmd_trans_huge_lock(pmdp, vma);
1478         if (ptl) {
1479                 u64 flags = 0, frame = 0;
1480                 pmd_t pmd = *pmdp;
1481                 struct page *page = NULL;
1482
1483                 if (vma->vm_flags & VM_SOFTDIRTY)
1484                         flags |= PM_SOFT_DIRTY;
1485
1486                 if (pmd_present(pmd)) {
1487                         page = pmd_page(pmd);
1488
1489                         flags |= PM_PRESENT;
1490                         if (pmd_soft_dirty(pmd))
1491                                 flags |= PM_SOFT_DIRTY;
1492                         if (pmd_uffd_wp(pmd))
1493                                 flags |= PM_UFFD_WP;
1494                         if (pm->show_pfn)
1495                                 frame = pmd_pfn(pmd) +
1496                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1497                 }
1498 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1499                 else if (is_swap_pmd(pmd)) {
1500                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1501                         unsigned long offset;
1502
1503                         if (pm->show_pfn) {
1504                                 if (is_pfn_swap_entry(entry))
1505                                         offset = swp_offset_pfn(entry);
1506                                 else
1507                                         offset = swp_offset(entry);
1508                                 offset = offset +
1509                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1510                                 frame = swp_type(entry) |
1511                                         (offset << MAX_SWAPFILES_SHIFT);
1512                         }
1513                         flags |= PM_SWAP;
1514                         if (pmd_swp_soft_dirty(pmd))
1515                                 flags |= PM_SOFT_DIRTY;
1516                         if (pmd_swp_uffd_wp(pmd))
1517                                 flags |= PM_UFFD_WP;
1518                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1519                         migration = is_migration_entry(entry);
1520                         page = pfn_swap_entry_to_page(entry);
1521                 }
1522 #endif
1523
1524                 if (page && !migration && page_mapcount(page) == 1)
1525                         flags |= PM_MMAP_EXCLUSIVE;
1526
1527                 for (; addr != end; addr += PAGE_SIZE) {
1528                         pagemap_entry_t pme = make_pme(frame, flags);
1529
1530                         err = add_to_pagemap(addr, &pme, pm);
1531                         if (err)
1532                                 break;
1533                         if (pm->show_pfn) {
1534                                 if (flags & PM_PRESENT)
1535                                         frame++;
1536                                 else if (flags & PM_SWAP)
1537                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1538                         }
1539                 }
1540                 spin_unlock(ptl);
1541                 return err;
1542         }
1543
1544         if (pmd_trans_unstable(pmdp))
1545                 return 0;
1546 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1547
1548         /*
1549          * We can assume that @vma always points to a valid one and @end never
1550          * goes beyond vma->vm_end.
1551          */
1552         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1553         for (; addr < end; pte++, addr += PAGE_SIZE) {
1554                 pagemap_entry_t pme;
1555
1556                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1557                 err = add_to_pagemap(addr, &pme, pm);
1558                 if (err)
1559                         break;
1560         }
1561         pte_unmap_unlock(orig_pte, ptl);
1562
1563         cond_resched();
1564
1565         return err;
1566 }
1567
1568 #ifdef CONFIG_HUGETLB_PAGE
1569 /* This function walks within one hugetlb entry in the single call */
1570 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1571                                  unsigned long addr, unsigned long end,
1572                                  struct mm_walk *walk)
1573 {
1574         struct pagemapread *pm = walk->private;
1575         struct vm_area_struct *vma = walk->vma;
1576         u64 flags = 0, frame = 0;
1577         int err = 0;
1578         pte_t pte;
1579
1580         if (vma->vm_flags & VM_SOFTDIRTY)
1581                 flags |= PM_SOFT_DIRTY;
1582
1583         pte = huge_ptep_get(ptep);
1584         if (pte_present(pte)) {
1585                 struct page *page = pte_page(pte);
1586
1587                 if (!PageAnon(page))
1588                         flags |= PM_FILE;
1589
1590                 if (page_mapcount(page) == 1)
1591                         flags |= PM_MMAP_EXCLUSIVE;
1592
1593                 if (huge_pte_uffd_wp(pte))
1594                         flags |= PM_UFFD_WP;
1595
1596                 flags |= PM_PRESENT;
1597                 if (pm->show_pfn)
1598                         frame = pte_pfn(pte) +
1599                                 ((addr & ~hmask) >> PAGE_SHIFT);
1600         } else if (pte_swp_uffd_wp_any(pte)) {
1601                 flags |= PM_UFFD_WP;
1602         }
1603
1604         for (; addr != end; addr += PAGE_SIZE) {
1605                 pagemap_entry_t pme = make_pme(frame, flags);
1606
1607                 err = add_to_pagemap(addr, &pme, pm);
1608                 if (err)
1609                         return err;
1610                 if (pm->show_pfn && (flags & PM_PRESENT))
1611                         frame++;
1612         }
1613
1614         cond_resched();
1615
1616         return err;
1617 }
1618 #else
1619 #define pagemap_hugetlb_range   NULL
1620 #endif /* HUGETLB_PAGE */
1621
1622 static const struct mm_walk_ops pagemap_ops = {
1623         .pmd_entry      = pagemap_pmd_range,
1624         .pte_hole       = pagemap_pte_hole,
1625         .hugetlb_entry  = pagemap_hugetlb_range,
1626 };
1627
1628 /*
1629  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1630  *
1631  * For each page in the address space, this file contains one 64-bit entry
1632  * consisting of the following:
1633  *
1634  * Bits 0-54  page frame number (PFN) if present
1635  * Bits 0-4   swap type if swapped
1636  * Bits 5-54  swap offset if swapped
1637  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1638  * Bit  56    page exclusively mapped
1639  * Bit  57    pte is uffd-wp write-protected
1640  * Bits 58-60 zero
1641  * Bit  61    page is file-page or shared-anon
1642  * Bit  62    page swapped
1643  * Bit  63    page present
1644  *
1645  * If the page is not present but in swap, then the PFN contains an
1646  * encoding of the swap file number and the page's offset into the
1647  * swap. Unmapped pages return a null PFN. This allows determining
1648  * precisely which pages are mapped (or in swap) and comparing mapped
1649  * pages between processes.
1650  *
1651  * Efficient users of this interface will use /proc/pid/maps to
1652  * determine which areas of memory are actually mapped and llseek to
1653  * skip over unmapped regions.
1654  */
1655 static ssize_t pagemap_read(struct file *file, char __user *buf,
1656                             size_t count, loff_t *ppos)
1657 {
1658         struct mm_struct *mm = file->private_data;
1659         struct pagemapread pm;
1660         unsigned long src;
1661         unsigned long svpfn;
1662         unsigned long start_vaddr;
1663         unsigned long end_vaddr;
1664         int ret = 0, copied = 0;
1665
1666         if (!mm || !mmget_not_zero(mm))
1667                 goto out;
1668
1669         ret = -EINVAL;
1670         /* file position must be aligned */
1671         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1672                 goto out_mm;
1673
1674         ret = 0;
1675         if (!count)
1676                 goto out_mm;
1677
1678         /* do not disclose physical addresses: attack vector */
1679         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1680
1681         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1682         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1683         ret = -ENOMEM;
1684         if (!pm.buffer)
1685                 goto out_mm;
1686
1687         src = *ppos;
1688         svpfn = src / PM_ENTRY_BYTES;
1689         end_vaddr = mm->task_size;
1690
1691         /* watch out for wraparound */
1692         start_vaddr = end_vaddr;
1693         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT))
1694                 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
1695
1696         /* Ensure the address is inside the task */
1697         if (start_vaddr > mm->task_size)
1698                 start_vaddr = end_vaddr;
1699
1700         /*
1701          * The odds are that this will stop walking way
1702          * before end_vaddr, because the length of the
1703          * user buffer is tracked in "pm", and the walk
1704          * will stop when we hit the end of the buffer.
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
1888         if (pmd_trans_unstable(pmd))
1889                 return 0;
1890 #endif
1891         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1892         do {
1893                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1894                 if (!page)
1895                         continue;
1896                 gather_stats(page, md, pte_dirty(*pte), 1);
1897
1898         } while (pte++, addr += PAGE_SIZE, addr != end);
1899         pte_unmap_unlock(orig_pte, ptl);
1900         cond_resched();
1901         return 0;
1902 }
1903 #ifdef CONFIG_HUGETLB_PAGE
1904 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1905                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1906 {
1907         pte_t huge_pte = huge_ptep_get(pte);
1908         struct numa_maps *md;
1909         struct page *page;
1910
1911         if (!pte_present(huge_pte))
1912                 return 0;
1913
1914         page = pte_page(huge_pte);
1915
1916         md = walk->private;
1917         gather_stats(page, md, pte_dirty(huge_pte), 1);
1918         return 0;
1919 }
1920
1921 #else
1922 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1923                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1924 {
1925         return 0;
1926 }
1927 #endif
1928
1929 static const struct mm_walk_ops show_numa_ops = {
1930         .hugetlb_entry = gather_hugetlb_stats,
1931         .pmd_entry = gather_pte_stats,
1932 };
1933
1934 /*
1935  * Display pages allocated per node and memory policy via /proc.
1936  */
1937 static int show_numa_map(struct seq_file *m, void *v)
1938 {
1939         struct numa_maps_private *numa_priv = m->private;
1940         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1941         struct vm_area_struct *vma = v;
1942         struct numa_maps *md = &numa_priv->md;
1943         struct file *file = vma->vm_file;
1944         struct mm_struct *mm = vma->vm_mm;
1945         struct mempolicy *pol;
1946         char buffer[64];
1947         int nid;
1948
1949         if (!mm)
1950                 return 0;
1951
1952         /* Ensure we start with an empty set of numa_maps statistics. */
1953         memset(md, 0, sizeof(*md));
1954
1955         pol = __get_vma_policy(vma, vma->vm_start);
1956         if (pol) {
1957                 mpol_to_str(buffer, sizeof(buffer), pol);
1958                 mpol_cond_put(pol);
1959         } else {
1960                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1961         }
1962
1963         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1964
1965         if (file) {
1966                 seq_puts(m, " file=");
1967                 seq_file_path(m, file, "\n\t= ");
1968         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1969                 seq_puts(m, " heap");
1970         } else if (is_stack(vma)) {
1971                 seq_puts(m, " stack");
1972         }
1973
1974         if (is_vm_hugetlb_page(vma))
1975                 seq_puts(m, " huge");
1976
1977         /* mmap_lock is held by m_start */
1978         walk_page_vma(vma, &show_numa_ops, md);
1979
1980         if (!md->pages)
1981                 goto out;
1982
1983         if (md->anon)
1984                 seq_printf(m, " anon=%lu", md->anon);
1985
1986         if (md->dirty)
1987                 seq_printf(m, " dirty=%lu", md->dirty);
1988
1989         if (md->pages != md->anon && md->pages != md->dirty)
1990                 seq_printf(m, " mapped=%lu", md->pages);
1991
1992         if (md->mapcount_max > 1)
1993                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1994
1995         if (md->swapcache)
1996                 seq_printf(m, " swapcache=%lu", md->swapcache);
1997
1998         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1999                 seq_printf(m, " active=%lu", md->active);
2000
2001         if (md->writeback)
2002                 seq_printf(m, " writeback=%lu", md->writeback);
2003
2004         for_each_node_state(nid, N_MEMORY)
2005                 if (md->node[nid])
2006                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2007
2008         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2009 out:
2010         seq_putc(m, '\n');
2011         return 0;
2012 }
2013
2014 static const struct seq_operations proc_pid_numa_maps_op = {
2015         .start  = m_start,
2016         .next   = m_next,
2017         .stop   = m_stop,
2018         .show   = show_numa_map,
2019 };
2020
2021 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2022 {
2023         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2024                                 sizeof(struct numa_maps_private));
2025 }
2026
2027 const struct file_operations proc_pid_numa_maps_operations = {
2028         .open           = pid_numa_maps_open,
2029         .read           = seq_read,
2030         .llseek         = seq_lseek,
2031         .release        = proc_map_release,
2032 };
2033
2034 #endif /* CONFIG_NUMA */