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         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 786                 /*
 787                  * For shared or readonly shmem mappings we know that all
 788                  * swapped out pages belong to the shmem object, and we can
 789                  * obtain the swap value much more efficiently. For private
 790                  * writable mappings, we might have COW pages that are
 791                  * not affected by the parent swapped out pages of the shmem
 792                  * object, so we have to distinguish them during the page walk.
 793                  * Unless we know that the shmem object (or the part mapped by
 794                  * our VMA) has no swapped out pages at all.
 795                  */
 796                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 797
 798                 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 799                                         !(vma->vm_flags & VM_WRITE))) {
 800                         mss->swap += shmem_swapped;
 801                 } else {
 802                         ops = &smaps_shmem_walk_ops;
 803                 }
 804         }
 805
 806         /* mmap_lock is held in m_start */
 807         if (!start)
 808                 walk_page_vma(vma, ops, mss);
 809         else
 810                 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
 811 }
 812
 813 #define SEQ_PUT_DEC(str, val) \
 814                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 815
 816 /* Show the contents common for smaps and smaps_rollup */
 817 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 818         bool rollup_mode)
 819 {
 820         SEQ_PUT_DEC("Rss:            ", mss->resident);
 821         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 822         SEQ_PUT_DEC(" kB\nPss_Dirty:      ", mss->pss_dirty >> PSS_SHIFT);
 823         if (rollup_mode) {
 824                 /*
 825                  * These are meaningful only for smaps_rollup, otherwise two of
 826                  * them are zero, and the other one is the same as Pss.
 827                  */
 828                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 829                         mss->pss_anon >> PSS_SHIFT);
 830                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 831                         mss->pss_file >> PSS_SHIFT);
 832                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 833                         mss->pss_shmem >> PSS_SHIFT);
 834         }
 835         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 836         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 837         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 838         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 839         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 840         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 841         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 842         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 843         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 844         SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
 845         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 846         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 847                                   mss->private_hugetlb >> 10, 7);
 848         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 849         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 850                                         mss->swap_pss >> PSS_SHIFT);
 851         SEQ_PUT_DEC(" kB\nLocked:         ",
 852                                         mss->pss_locked >> PSS_SHIFT);
 853         seq_puts(m, " kB\n");
 854 }
 855
 856 static int show_smap(struct seq_file *m, void *v)
 857 {
 858         struct vm_area_struct *vma = v;
 859         struct mem_size_stats mss;
 860
 861         memset(&mss, 0, sizeof(mss));
 862
 863         smap_gather_stats(vma, &mss, 0);
 864
 865         show_map_vma(m, vma);
 866
 867         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 868         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 869         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 870         seq_puts(m, " kB\n");
 871
 872         __show_smap(m, &mss, false);
 873
 874         seq_printf(m, "THPeligible:    %d\n",
 875                    hugepage_vma_check(vma, vma->vm_flags, true, false, true));
 876
 877         if (arch_pkeys_enabled())
 878                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 879         show_smap_vma_flags(m, vma);
 880
 881         return 0;
 882 }
 883
 884 static int show_smaps_rollup(struct seq_file *m, void *v)
 885 {
 886         struct proc_maps_private *priv = m->private;
 887         struct mem_size_stats mss;
 888         struct mm_struct *mm = priv->mm;
 889         struct vm_area_struct *vma;
 890         unsigned long vma_start = 0, last_vma_end = 0;
 891         int ret = 0;
 892         VMA_ITERATOR(vmi, mm, 0);
 893
 894         priv->task = get_proc_task(priv->inode);
 895         if (!priv->task)
 896                 return -ESRCH;
 897
 898         if (!mm || !mmget_not_zero(mm)) {
 899                 ret = -ESRCH;
 900                 goto out_put_task;
 901         }
 902
 903         memset(&mss, 0, sizeof(mss));
 904
 905         ret = mmap_read_lock_killable(mm);
 906         if (ret)
 907                 goto out_put_mm;
 908
 909         hold_task_mempolicy(priv);
 910         vma = vma_next(&vmi);
 911
 912         if (unlikely(!vma))
 913                 goto empty_set;
 914
 915         vma_start = vma->vm_start;
 916         do {
 917                 smap_gather_stats(vma, &mss, 0);
 918                 last_vma_end = vma->vm_end;
 919
 920                 /*
 921                  * Release mmap_lock temporarily if someone wants to
 922                  * access it for write request.
 923                  */
 924                 if (mmap_lock_is_contended(mm)) {
 925                         vma_iter_invalidate(&vmi);
 926                         mmap_read_unlock(mm);
 927                         ret = mmap_read_lock_killable(mm);
 928                         if (ret) {
 929                                 release_task_mempolicy(priv);
 930                                 goto out_put_mm;
 931                         }
 932
 933                         /*
 934                          * After dropping the lock, there are four cases to
 935                          * consider. See the following example for explanation.
 936                          *
 937                          *   +------+------+-----------+
 938                          *   | VMA1 | VMA2 | VMA3      |
 939                          *   +------+------+-----------+
 940                          *   |      |      |           |
 941                          *  4k     8k     16k         400k
 942                          *
 943                          * Suppose we drop the lock after reading VMA2 due to
 944                          * contention, then we get:
 945                          *
 946                          *      last_vma_end = 16k
 947                          *
 948                          * 1) VMA2 is freed, but VMA3 exists:
 949                          *
 950                          *    vma_next(vmi) will return VMA3.
 951                          *    In this case, just continue from VMA3.
 952                          *
 953                          * 2) VMA2 still exists:
 954                          *
 955                          *    vma_next(vmi) will return VMA3.
 956                          *    In this case, just continue from VMA3.
 957                          *
 958                          * 3) No more VMAs can be found:
 959                          *
 960                          *    vma_next(vmi) will return NULL.
 961                          *    No more things to do, just break.
 962                          *
 963                          * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
 964                          *
 965                          *    vma_next(vmi) will return VMA' whose range
 966                          *    contains last_vma_end.
 967                          *    Iterate VMA' from last_vma_end.
 968                          */
 969                         vma = vma_next(&vmi);
 970                         /* Case 3 above */
 971                         if (!vma)
 972                                 break;
 973
 974                         /* Case 1 and 2 above */
 975                         if (vma->vm_start >= last_vma_end)
 976                                 continue;
 977
 978                         /* Case 4 above */
 979                         if (vma->vm_end > last_vma_end)
 980                                 smap_gather_stats(vma, &mss, last_vma_end);
 981                 }
 982         } for_each_vma(vmi, vma);
 983
 984 empty_set:
 985         show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
 986         seq_pad(m, ' ');
 987         seq_puts(m, "[rollup]\n");
 988
 989         __show_smap(m, &mss, true);
 990
 991         release_task_mempolicy(priv);
 992         mmap_read_unlock(mm);
 993
 994 out_put_mm:
 995         mmput(mm);
 996 out_put_task:
 997         put_task_struct(priv->task);
 998         priv->task = NULL;
 999
1000         return ret;
1001 }
1002 #undef SEQ_PUT_DEC
1003
1004 static const struct seq_operations proc_pid_smaps_op = {
1005         .start  = m_start,
1006         .next   = m_next,
1007         .stop   = m_stop,
1008         .show   = show_smap
1009 };
1010
1011 static int pid_smaps_open(struct inode *inode, struct file *file)
1012 {
1013         return do_maps_open(inode, file, &proc_pid_smaps_op);
1014 }
1015
1016 static int smaps_rollup_open(struct inode *inode, struct file *file)
1017 {
1018         int ret;
1019         struct proc_maps_private *priv;
1020
1021         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1022         if (!priv)
1023                 return -ENOMEM;
1024
1025         ret = single_open(file, show_smaps_rollup, priv);
1026         if (ret)
1027                 goto out_free;
1028
1029         priv->inode = inode;
1030         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1031         if (IS_ERR(priv->mm)) {
1032                 ret = PTR_ERR(priv->mm);
1033
1034                 single_release(inode, file);
1035                 goto out_free;
1036         }
1037
1038         return 0;
1039
1040 out_free:
1041         kfree(priv);
1042         return ret;
1043 }
1044
1045 static int smaps_rollup_release(struct inode *inode, struct file *file)
1046 {
1047         struct seq_file *seq = file->private_data;
1048         struct proc_maps_private *priv = seq->private;
1049
1050         if (priv->mm)
1051                 mmdrop(priv->mm);
1052
1053         kfree(priv);
1054         return single_release(inode, file);
1055 }
1056
1057 const struct file_operations proc_pid_smaps_operations = {
1058         .open           = pid_smaps_open,
1059         .read           = seq_read,
1060         .llseek         = seq_lseek,
1061         .release        = proc_map_release,
1062 };
1063
1064 const struct file_operations proc_pid_smaps_rollup_operations = {
1065         .open           = smaps_rollup_open,
1066         .read           = seq_read,
1067         .llseek         = seq_lseek,
1068         .release        = smaps_rollup_release,
1069 };
1070
1071 enum clear_refs_types {
1072         CLEAR_REFS_ALL = 1,
1073         CLEAR_REFS_ANON,
1074         CLEAR_REFS_MAPPED,
1075         CLEAR_REFS_SOFT_DIRTY,
1076         CLEAR_REFS_MM_HIWATER_RSS,
1077         CLEAR_REFS_LAST,
1078 };
1079
1080 struct clear_refs_private {
1081         enum clear_refs_types type;
1082 };
1083
1084 #ifdef CONFIG_MEM_SOFT_DIRTY
1085
1086 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1087 {
1088         struct page *page;
1089
1090         if (!pte_write(pte))
1091                 return false;
1092         if (!is_cow_mapping(vma->vm_flags))
1093                 return false;
1094         if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1095                 return false;
1096         page = vm_normal_page(vma, addr, pte);
1097         if (!page)
1098                 return false;
1099         return page_maybe_dma_pinned(page);
1100 }
1101
1102 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1103                 unsigned long addr, pte_t *pte)
1104 {
1105         /*
1106          * The soft-dirty tracker uses #PF-s to catch writes
1107          * to pages, so write-protect the pte as well. See the
1108          * Documentation/admin-guide/mm/soft-dirty.rst for full description
1109          * of how soft-dirty works.
1110          */
1111         pte_t ptent = *pte;
1112
1113         if (pte_present(ptent)) {
1114                 pte_t old_pte;
1115
1116                 if (pte_is_pinned(vma, addr, ptent))
1117                         return;
1118                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1119                 ptent = pte_wrprotect(old_pte);
1120                 ptent = pte_clear_soft_dirty(ptent);
1121                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1122         } else if (is_swap_pte(ptent)) {
1123                 ptent = pte_swp_clear_soft_dirty(ptent);
1124                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1125         }
1126 }
1127 #else
1128 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1129                 unsigned long addr, pte_t *pte)
1130 {
1131 }
1132 #endif
1133
1134 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1135 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1136                 unsigned long addr, pmd_t *pmdp)
1137 {
1138         pmd_t old, pmd = *pmdp;
1139
1140         if (pmd_present(pmd)) {
1141                 /* See comment in change_huge_pmd() */
1142                 old = pmdp_invalidate(vma, addr, pmdp);
1143                 if (pmd_dirty(old))
1144                         pmd = pmd_mkdirty(pmd);
1145                 if (pmd_young(old))
1146                         pmd = pmd_mkyoung(pmd);
1147
1148                 pmd = pmd_wrprotect(pmd);
1149                 pmd = pmd_clear_soft_dirty(pmd);
1150
1151                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1152         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1153                 pmd = pmd_swp_clear_soft_dirty(pmd);
1154                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1155         }
1156 }
1157 #else
1158 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1159                 unsigned long addr, pmd_t *pmdp)
1160 {
1161 }
1162 #endif
1163
1164 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1165                                 unsigned long end, struct mm_walk *walk)
1166 {
1167         struct clear_refs_private *cp = walk->private;
1168         struct vm_area_struct *vma = walk->vma;
1169         pte_t *pte, ptent;
1170         spinlock_t *ptl;
1171         struct page *page;
1172
1173         ptl = pmd_trans_huge_lock(pmd, vma);
1174         if (ptl) {
1175                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1176                         clear_soft_dirty_pmd(vma, addr, pmd);
1177                         goto out;
1178                 }
1179
1180                 if (!pmd_present(*pmd))
1181                         goto out;
1182
1183                 page = pmd_page(*pmd);
1184
1185                 /* Clear accessed and referenced bits. */
1186                 pmdp_test_and_clear_young(vma, addr, pmd);
1187                 test_and_clear_page_young(page);
1188                 ClearPageReferenced(page);
1189 out:
1190                 spin_unlock(ptl);
1191                 return 0;
1192         }
1193
1194         if (pmd_trans_unstable(pmd))
1195                 return 0;
1196
1197         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1198         for (; addr != end; pte++, addr += PAGE_SIZE) {
1199                 ptent = *pte;
1200
1201                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1202                         clear_soft_dirty(vma, addr, pte);
1203                         continue;
1204                 }
1205
1206                 if (!pte_present(ptent))
1207                         continue;
1208
1209                 page = vm_normal_page(vma, addr, ptent);
1210                 if (!page)
1211                         continue;
1212
1213                 /* Clear accessed and referenced bits. */
1214                 ptep_test_and_clear_young(vma, addr, pte);
1215                 test_and_clear_page_young(page);
1216                 ClearPageReferenced(page);
1217         }
1218         pte_unmap_unlock(pte - 1, ptl);
1219         cond_resched();
1220         return 0;
1221 }
1222
1223 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1224                                 struct mm_walk *walk)
1225 {
1226         struct clear_refs_private *cp = walk->private;
1227         struct vm_area_struct *vma = walk->vma;
1228
1229         if (vma->vm_flags & VM_PFNMAP)
1230                 return 1;
1231
1232         /*
1233          * Writing 1 to /proc/pid/clear_refs affects all pages.
1234          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1235          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1236          * Writing 4 to /proc/pid/clear_refs affects all pages.
1237          */
1238         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1239                 return 1;
1240         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1241                 return 1;
1242         return 0;
1243 }
1244
1245 static const struct mm_walk_ops clear_refs_walk_ops = {
1246         .pmd_entry              = clear_refs_pte_range,
1247         .test_walk              = clear_refs_test_walk,
1248 };
1249
1250 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1251                                 size_t count, loff_t *ppos)
1252 {
1253         struct task_struct *task;
1254         char buffer[PROC_NUMBUF];
1255         struct mm_struct *mm;
1256         struct vm_area_struct *vma;
1257         enum clear_refs_types type;
1258         int itype;
1259         int rv;
1260
1261         memset(buffer, 0, sizeof(buffer));
1262         if (count > sizeof(buffer) - 1)
1263                 count = sizeof(buffer) - 1;
1264         if (copy_from_user(buffer, buf, count))
1265                 return -EFAULT;
1266         rv = kstrtoint(strstrip(buffer), 10, &itype);
1267         if (rv < 0)
1268                 return rv;
1269         type = (enum clear_refs_types)itype;
1270         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1271                 return -EINVAL;
1272
1273         task = get_proc_task(file_inode(file));
1274         if (!task)
1275                 return -ESRCH;
1276         mm = get_task_mm(task);
1277         if (mm) {
1278                 VMA_ITERATOR(vmi, mm, 0);
1279                 struct mmu_notifier_range range;
1280                 struct clear_refs_private cp = {
1281                         .type = type,
1282                 };
1283
1284                 if (mmap_write_lock_killable(mm)) {
1285                         count = -EINTR;
1286                         goto out_mm;
1287                 }
1288                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1289                         /*
1290                          * Writing 5 to /proc/pid/clear_refs resets the peak
1291                          * resident set size to this mm's current rss value.
1292                          */
1293                         reset_mm_hiwater_rss(mm);
1294                         goto out_unlock;
1295                 }
1296
1297                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1298                         for_each_vma(vmi, vma) {
1299                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1300                                         continue;
1301                                 vm_flags_clear(vma, VM_SOFTDIRTY);
1302                                 vma_set_page_prot(vma);
1303                         }
1304
1305                         inc_tlb_flush_pending(mm);
1306                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1307                                                 0, mm, 0, -1UL);
1308                         mmu_notifier_invalidate_range_start(&range);
1309                 }
1310                 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1311                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1312                         mmu_notifier_invalidate_range_end(&range);
1313                         flush_tlb_mm(mm);
1314                         dec_tlb_flush_pending(mm);
1315                 }
1316 out_unlock:
1317                 mmap_write_unlock(mm);
1318 out_mm:
1319                 mmput(mm);
1320         }
1321         put_task_struct(task);
1322
1323         return count;
1324 }
1325
1326 const struct file_operations proc_clear_refs_operations = {
1327         .write          = clear_refs_write,
1328         .llseek         = noop_llseek,
1329 };
1330
1331 typedef struct {
1332         u64 pme;
1333 } pagemap_entry_t;
1334
1335 struct pagemapread {
1336         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1337         pagemap_entry_t *buffer;
1338         bool show_pfn;
1339 };
1340
1341 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1342 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1343
1344 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1345 #define PM_PFRAME_BITS          55
1346 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1347 #define PM_SOFT_DIRTY           BIT_ULL(55)
1348 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1349 #define PM_UFFD_WP              BIT_ULL(57)
1350 #define PM_FILE                 BIT_ULL(61)
1351 #define PM_SWAP                 BIT_ULL(62)
1352 #define PM_PRESENT              BIT_ULL(63)
1353
1354 #define PM_END_OF_BUFFER    1
1355
1356 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1357 {
1358         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1359 }
1360
1361 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1362                           struct pagemapread *pm)
1363 {
1364         pm->buffer[pm->pos++] = *pme;
1365         if (pm->pos >= pm->len)
1366                 return PM_END_OF_BUFFER;
1367         return 0;
1368 }
1369
1370 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1371                             __always_unused int depth, struct mm_walk *walk)
1372 {
1373         struct pagemapread *pm = walk->private;
1374         unsigned long addr = start;
1375         int err = 0;
1376
1377         while (addr < end) {
1378                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1379                 pagemap_entry_t pme = make_pme(0, 0);
1380                 /* End of address space hole, which we mark as non-present. */
1381                 unsigned long hole_end;
1382
1383                 if (vma)
1384                         hole_end = min(end, vma->vm_start);
1385                 else
1386                         hole_end = end;
1387
1388                 for (; addr < hole_end; addr += PAGE_SIZE) {
1389                         err = add_to_pagemap(addr, &pme, pm);
1390                         if (err)
1391                                 goto out;
1392                 }
1393
1394                 if (!vma)
1395                         break;
1396
1397                 /* Addresses in the VMA. */
1398                 if (vma->vm_flags & VM_SOFTDIRTY)
1399                         pme = make_pme(0, PM_SOFT_DIRTY);
1400                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1401                         err = add_to_pagemap(addr, &pme, pm);
1402                         if (err)
1403                                 goto out;
1404                 }
1405         }
1406 out:
1407         return err;
1408 }
1409
1410 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1411                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1412 {
1413         u64 frame = 0, flags = 0;
1414         struct page *page = NULL;
1415         bool migration = false;
1416
1417         if (pte_present(pte)) {
1418                 if (pm->show_pfn)
1419                         frame = pte_pfn(pte);
1420                 flags |= PM_PRESENT;
1421                 page = vm_normal_page(vma, addr, pte);
1422                 if (pte_soft_dirty(pte))
1423                         flags |= PM_SOFT_DIRTY;
1424                 if (pte_uffd_wp(pte))
1425                         flags |= PM_UFFD_WP;
1426         } else if (is_swap_pte(pte)) {
1427                 swp_entry_t entry;
1428                 if (pte_swp_soft_dirty(pte))
1429                         flags |= PM_SOFT_DIRTY;
1430                 if (pte_swp_uffd_wp(pte))
1431                         flags |= PM_UFFD_WP;
1432                 entry = pte_to_swp_entry(pte);
1433                 if (pm->show_pfn) {
1434                         pgoff_t offset;
1435                         /*
1436                          * For PFN swap offsets, keeping the offset field
1437                          * to be PFN only to be compatible with old smaps.
1438                          */
1439                         if (is_pfn_swap_entry(entry))
1440                                 offset = swp_offset_pfn(entry);
1441                         else
1442                                 offset = swp_offset(entry);
1443                         frame = swp_type(entry) |
1444                             (offset << MAX_SWAPFILES_SHIFT);
1445                 }
1446                 flags |= PM_SWAP;
1447                 migration = is_migration_entry(entry);
1448                 if (is_pfn_swap_entry(entry))
1449                         page = pfn_swap_entry_to_page(entry);
1450                 if (pte_marker_entry_uffd_wp(entry))
1451                         flags |= PM_UFFD_WP;
1452         }
1453
1454         if (page && !PageAnon(page))
1455                 flags |= PM_FILE;
1456         if (page && !migration && page_mapcount(page) == 1)
1457                 flags |= PM_MMAP_EXCLUSIVE;
1458         if (vma->vm_flags & VM_SOFTDIRTY)
1459                 flags |= PM_SOFT_DIRTY;
1460
1461         return make_pme(frame, flags);
1462 }
1463
1464 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1465                              struct mm_walk *walk)
1466 {
1467         struct vm_area_struct *vma = walk->vma;
1468         struct pagemapread *pm = walk->private;
1469         spinlock_t *ptl;
1470         pte_t *pte, *orig_pte;
1471         int err = 0;
1472 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1473         bool migration = false;
1474
1475         ptl = pmd_trans_huge_lock(pmdp, vma);
1476         if (ptl) {
1477                 u64 flags = 0, frame = 0;
1478                 pmd_t pmd = *pmdp;
1479                 struct page *page = NULL;
1480
1481                 if (vma->vm_flags & VM_SOFTDIRTY)
1482                         flags |= PM_SOFT_DIRTY;
1483
1484                 if (pmd_present(pmd)) {
1485                         page = pmd_page(pmd);
1486
1487                         flags |= PM_PRESENT;
1488                         if (pmd_soft_dirty(pmd))
1489                                 flags |= PM_SOFT_DIRTY;
1490                         if (pmd_uffd_wp(pmd))
1491                                 flags |= PM_UFFD_WP;
1492                         if (pm->show_pfn)
1493                                 frame = pmd_pfn(pmd) +
1494                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1495                 }
1496 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1497                 else if (is_swap_pmd(pmd)) {
1498                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1499                         unsigned long offset;
1500
1501                         if (pm->show_pfn) {
1502                                 if (is_pfn_swap_entry(entry))
1503                                         offset = swp_offset_pfn(entry);
1504                                 else
1505                                         offset = swp_offset(entry);
1506                                 offset = offset +
1507                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1508                                 frame = swp_type(entry) |
1509                                         (offset << MAX_SWAPFILES_SHIFT);
1510                         }
1511                         flags |= PM_SWAP;
1512                         if (pmd_swp_soft_dirty(pmd))
1513                                 flags |= PM_SOFT_DIRTY;
1514                         if (pmd_swp_uffd_wp(pmd))
1515                                 flags |= PM_UFFD_WP;
1516                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1517                         migration = is_migration_entry(entry);
1518                         page = pfn_swap_entry_to_page(entry);
1519                 }
1520 #endif
1521
1522                 if (page && !migration && page_mapcount(page) == 1)
1523                         flags |= PM_MMAP_EXCLUSIVE;
1524
1525                 for (; addr != end; addr += PAGE_SIZE) {
1526                         pagemap_entry_t pme = make_pme(frame, flags);
1527
1528                         err = add_to_pagemap(addr, &pme, pm);
1529                         if (err)
1530                                 break;
1531                         if (pm->show_pfn) {
1532                                 if (flags & PM_PRESENT)
1533                                         frame++;
1534                                 else if (flags & PM_SWAP)
1535                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1536                         }
1537                 }
1538                 spin_unlock(ptl);
1539                 return err;
1540         }
1541
1542         if (pmd_trans_unstable(pmdp))
1543                 return 0;
1544 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1545
1546         /*
1547          * We can assume that @vma always points to a valid one and @end never
1548          * goes beyond vma->vm_end.
1549          */
1550         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1551         for (; addr < end; pte++, addr += PAGE_SIZE) {
1552                 pagemap_entry_t pme;
1553
1554                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1555                 err = add_to_pagemap(addr, &pme, pm);
1556                 if (err)
1557                         break;
1558         }
1559         pte_unmap_unlock(orig_pte, ptl);
1560
1561         cond_resched();
1562
1563         return err;
1564 }
1565
1566 #ifdef CONFIG_HUGETLB_PAGE
1567 /* This function walks within one hugetlb entry in the single call */
1568 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1569                                  unsigned long addr, unsigned long end,
1570                                  struct mm_walk *walk)
1571 {
1572         struct pagemapread *pm = walk->private;
1573         struct vm_area_struct *vma = walk->vma;
1574         u64 flags = 0, frame = 0;
1575         int err = 0;
1576         pte_t pte;
1577
1578         if (vma->vm_flags & VM_SOFTDIRTY)
1579                 flags |= PM_SOFT_DIRTY;
1580
1581         pte = huge_ptep_get(ptep);
1582         if (pte_present(pte)) {
1583                 struct page *page = pte_page(pte);
1584
1585                 if (!PageAnon(page))
1586                         flags |= PM_FILE;
1587
1588                 if (page_mapcount(page) == 1)
1589                         flags |= PM_MMAP_EXCLUSIVE;
1590
1591                 if (huge_pte_uffd_wp(pte))
1592                         flags |= PM_UFFD_WP;
1593
1594                 flags |= PM_PRESENT;
1595                 if (pm->show_pfn)
1596                         frame = pte_pfn(pte) +
1597                                 ((addr & ~hmask) >> PAGE_SHIFT);
1598         } else if (pte_swp_uffd_wp_any(pte)) {
1599                 flags |= PM_UFFD_WP;
1600         }
1601
1602         for (; addr != end; addr += PAGE_SIZE) {
1603                 pagemap_entry_t pme = make_pme(frame, flags);
1604
1605                 err = add_to_pagemap(addr, &pme, pm);
1606                 if (err)
1607                         return err;
1608                 if (pm->show_pfn && (flags & PM_PRESENT))
1609                         frame++;
1610         }
1611
1612         cond_resched();
1613
1614         return err;
1615 }
1616 #else
1617 #define pagemap_hugetlb_range   NULL
1618 #endif /* HUGETLB_PAGE */
1619
1620 static const struct mm_walk_ops pagemap_ops = {
1621         .pmd_entry      = pagemap_pmd_range,
1622         .pte_hole       = pagemap_pte_hole,
1623         .hugetlb_entry  = pagemap_hugetlb_range,
1624 };
1625
1626 /*
1627  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1628  *
1629  * For each page in the address space, this file contains one 64-bit entry
1630  * consisting of the following:
1631  *
1632  * Bits 0-54  page frame number (PFN) if present
1633  * Bits 0-4   swap type if swapped
1634  * Bits 5-54  swap offset if swapped
1635  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1636  * Bit  56    page exclusively mapped
1637  * Bit  57    pte is uffd-wp write-protected
1638  * Bits 58-60 zero
1639  * Bit  61    page is file-page or shared-anon
1640  * Bit  62    page swapped
1641  * Bit  63    page present
1642  *
1643  * If the page is not present but in swap, then the PFN contains an
1644  * encoding of the swap file number and the page's offset into the
1645  * swap. Unmapped pages return a null PFN. This allows determining
1646  * precisely which pages are mapped (or in swap) and comparing mapped
1647  * pages between processes.
1648  *
1649  * Efficient users of this interface will use /proc/pid/maps to
1650  * determine which areas of memory are actually mapped and llseek to
1651  * skip over unmapped regions.
1652  */
1653 static ssize_t pagemap_read(struct file *file, char __user *buf,
1654                             size_t count, loff_t *ppos)
1655 {
1656         struct mm_struct *mm = file->private_data;
1657         struct pagemapread pm;
1658         unsigned long src;
1659         unsigned long svpfn;
1660         unsigned long start_vaddr;
1661         unsigned long end_vaddr;
1662         int ret = 0, copied = 0;
1663
1664         if (!mm || !mmget_not_zero(mm))
1665                 goto out;
1666
1667         ret = -EINVAL;
1668         /* file position must be aligned */
1669         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1670                 goto out_mm;
1671
1672         ret = 0;
1673         if (!count)
1674                 goto out_mm;
1675
1676         /* do not disclose physical addresses: attack vector */
1677         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1678
1679         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1680         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1681         ret = -ENOMEM;
1682         if (!pm.buffer)
1683                 goto out_mm;
1684
1685         src = *ppos;
1686         svpfn = src / PM_ENTRY_BYTES;
1687         end_vaddr = mm->task_size;
1688
1689         /* watch out for wraparound */
1690         start_vaddr = end_vaddr;
1691         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
1692                 ret = mmap_read_lock_killable(mm);
1693                 if (ret)
1694                         goto out_free;
1695                 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
1696                 mmap_read_unlock(mm);
1697         }
1698
1699         /* Ensure the address is inside the task */
1700         if (start_vaddr > mm->task_size)
1701                 start_vaddr = end_vaddr;
1702
1703         /*
1704          * The odds are that this will stop walking way
1705          * before end_vaddr, because the length of the
1706          * user buffer is tracked in "pm", and the walk
1707          * will stop when we hit the end of the buffer.
1708          */
1709         ret = 0;
1710         while (count && (start_vaddr < end_vaddr)) {
1711                 int len;
1712                 unsigned long end;
1713
1714                 pm.pos = 0;
1715                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1716                 /* overflow ? */
1717                 if (end < start_vaddr || end > end_vaddr)
1718                         end = end_vaddr;
1719                 ret = mmap_read_lock_killable(mm);
1720                 if (ret)
1721                         goto out_free;
1722                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1723                 mmap_read_unlock(mm);
1724                 start_vaddr = end;
1725
1726                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1727                 if (copy_to_user(buf, pm.buffer, len)) {
1728                         ret = -EFAULT;
1729                         goto out_free;
1730                 }
1731                 copied += len;
1732                 buf += len;
1733                 count -= len;
1734         }
1735         *ppos += copied;
1736         if (!ret || ret == PM_END_OF_BUFFER)
1737                 ret = copied;
1738
1739 out_free:
1740         kfree(pm.buffer);
1741 out_mm:
1742         mmput(mm);
1743 out:
1744         return ret;
1745 }
1746
1747 static int pagemap_open(struct inode *inode, struct file *file)
1748 {
1749         struct mm_struct *mm;
1750
1751         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1752         if (IS_ERR(mm))
1753                 return PTR_ERR(mm);
1754         file->private_data = mm;
1755         return 0;
1756 }
1757
1758 static int pagemap_release(struct inode *inode, struct file *file)
1759 {
1760         struct mm_struct *mm = file->private_data;
1761
1762         if (mm)
1763                 mmdrop(mm);
1764         return 0;
1765 }
1766
1767 const struct file_operations proc_pagemap_operations = {
1768         .llseek         = mem_lseek, /* borrow this */
1769         .read           = pagemap_read,
1770         .open           = pagemap_open,
1771         .release        = pagemap_release,
1772 };
1773 #endif /* CONFIG_PROC_PAGE_MONITOR */
1774
1775 #ifdef CONFIG_NUMA
1776
1777 struct numa_maps {
1778         unsigned long pages;
1779         unsigned long anon;
1780         unsigned long active;
1781         unsigned long writeback;
1782         unsigned long mapcount_max;
1783         unsigned long dirty;
1784         unsigned long swapcache;
1785         unsigned long node[MAX_NUMNODES];
1786 };
1787
1788 struct numa_maps_private {
1789         struct proc_maps_private proc_maps;
1790         struct numa_maps md;
1791 };
1792
1793 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1794                         unsigned long nr_pages)
1795 {
1796         int count = page_mapcount(page);
1797
1798         md->pages += nr_pages;
1799         if (pte_dirty || PageDirty(page))
1800                 md->dirty += nr_pages;
1801
1802         if (PageSwapCache(page))
1803                 md->swapcache += nr_pages;
1804
1805         if (PageActive(page) || PageUnevictable(page))
1806                 md->active += nr_pages;
1807
1808         if (PageWriteback(page))
1809                 md->writeback += nr_pages;
1810
1811         if (PageAnon(page))
1812                 md->anon += nr_pages;
1813
1814         if (count > md->mapcount_max)
1815                 md->mapcount_max = count;
1816
1817         md->node[page_to_nid(page)] += nr_pages;
1818 }
1819
1820 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1821                 unsigned long addr)
1822 {
1823         struct page *page;
1824         int nid;
1825
1826         if (!pte_present(pte))
1827                 return NULL;
1828
1829         page = vm_normal_page(vma, addr, pte);
1830         if (!page || is_zone_device_page(page))
1831                 return NULL;
1832
1833         if (PageReserved(page))
1834                 return NULL;
1835
1836         nid = page_to_nid(page);
1837         if (!node_isset(nid, node_states[N_MEMORY]))
1838                 return NULL;
1839
1840         return page;
1841 }
1842
1843 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1844 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1845                                               struct vm_area_struct *vma,
1846                                               unsigned long addr)
1847 {
1848         struct page *page;
1849         int nid;
1850
1851         if (!pmd_present(pmd))
1852                 return NULL;
1853
1854         page = vm_normal_page_pmd(vma, addr, pmd);
1855         if (!page)
1856                 return NULL;
1857
1858         if (PageReserved(page))
1859                 return NULL;
1860
1861         nid = page_to_nid(page);
1862         if (!node_isset(nid, node_states[N_MEMORY]))
1863                 return NULL;
1864
1865         return page;
1866 }
1867 #endif
1868
1869 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1870                 unsigned long end, struct mm_walk *walk)
1871 {
1872         struct numa_maps *md = walk->private;
1873         struct vm_area_struct *vma = walk->vma;
1874         spinlock_t *ptl;
1875         pte_t *orig_pte;
1876         pte_t *pte;
1877
1878 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1879         ptl = pmd_trans_huge_lock(pmd, vma);
1880         if (ptl) {
1881                 struct page *page;
1882
1883                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1884                 if (page)
1885                         gather_stats(page, md, pmd_dirty(*pmd),
1886                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1887                 spin_unlock(ptl);
1888                 return 0;
1889         }
1890
1891         if (pmd_trans_unstable(pmd))
1892                 return 0;
1893 #endif
1894         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1895         do {
1896                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1897                 if (!page)
1898                         continue;
1899                 gather_stats(page, md, pte_dirty(*pte), 1);
1900
1901         } while (pte++, addr += PAGE_SIZE, addr != end);
1902         pte_unmap_unlock(orig_pte, ptl);
1903         cond_resched();
1904         return 0;
1905 }
1906 #ifdef CONFIG_HUGETLB_PAGE
1907 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1908                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1909 {
1910         pte_t huge_pte = huge_ptep_get(pte);
1911         struct numa_maps *md;
1912         struct page *page;
1913
1914         if (!pte_present(huge_pte))
1915                 return 0;
1916
1917         page = pte_page(huge_pte);
1918
1919         md = walk->private;
1920         gather_stats(page, md, pte_dirty(huge_pte), 1);
1921         return 0;
1922 }
1923
1924 #else
1925 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1926                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1927 {
1928         return 0;
1929 }
1930 #endif
1931
1932 static const struct mm_walk_ops show_numa_ops = {
1933         .hugetlb_entry = gather_hugetlb_stats,
1934         .pmd_entry = gather_pte_stats,
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->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1972                 seq_puts(m, " heap");
1973         } else if (is_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 */