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         pte_t ptent = ptep_get(pte);
 542
 543         if (pte_present(ptent)) {
 544                 page = vm_normal_page(vma, addr, ptent);
 545                 young = pte_young(ptent);
 546                 dirty = pte_dirty(ptent);
 547         } else if (is_swap_pte(ptent)) {
 548                 swp_entry_t swpent = pte_to_swp_entry(ptent);
 549
 550                 if (!non_swap_entry(swpent)) {
 551                         int mapcount;
 552
 553                         mss->swap += PAGE_SIZE;
 554                         mapcount = swp_swapcount(swpent);
 555                         if (mapcount >= 2) {
 556                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 557
 558                                 do_div(pss_delta, mapcount);
 559                                 mss->swap_pss += pss_delta;
 560                         } else {
 561                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 562                         }
 563                 } else if (is_pfn_swap_entry(swpent)) {
 564                         if (is_migration_entry(swpent))
 565                                 migration = true;
 566                         page = pfn_swap_entry_to_page(swpent);
 567                 }
 568         } else {
 569                 smaps_pte_hole_lookup(addr, walk);
 570                 return;
 571         }
 572
 573         if (!page)
 574                 return;
 575
 576         smaps_account(mss, page, false, young, dirty, locked, migration);
 577 }
 578
 579 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 580 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 581                 struct mm_walk *walk)
 582 {
 583         struct mem_size_stats *mss = walk->private;
 584         struct vm_area_struct *vma = walk->vma;
 585         bool locked = !!(vma->vm_flags & VM_LOCKED);
 586         struct page *page = NULL;
 587         bool migration = false;
 588
 589         if (pmd_present(*pmd)) {
 590                 page = vm_normal_page_pmd(vma, addr, *pmd);
 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         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 635         if (!pte) {
 636                 walk->action = ACTION_AGAIN;
 637                 return 0;
 638         }
 639         for (; addr != end; pte++, addr += PAGE_SIZE)
 640                 smaps_pte_entry(pte, addr, walk);
 641         pte_unmap_unlock(pte - 1, ptl);
 642 out:
 643         cond_resched();
 644         return 0;
 645 }
 646
 647 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 648 {
 649         /*
 650          * Don't forget to update Documentation/ on changes.
 651          */
 652         static const char mnemonics[BITS_PER_LONG][2] = {
 653                 /*
 654                  * In case if we meet a flag we don't know about.
 655                  */
 656                 [0 ... (BITS_PER_LONG-1)] = "??",
 657
 658                 [ilog2(VM_READ)]        = "rd",
 659                 [ilog2(VM_WRITE)]       = "wr",
 660                 [ilog2(VM_EXEC)]        = "ex",
 661                 [ilog2(VM_SHARED)]      = "sh",
 662                 [ilog2(VM_MAYREAD)]     = "mr",
 663                 [ilog2(VM_MAYWRITE)]    = "mw",
 664                 [ilog2(VM_MAYEXEC)]     = "me",
 665                 [ilog2(VM_MAYSHARE)]    = "ms",
 666                 [ilog2(VM_GROWSDOWN)]   = "gd",
 667                 [ilog2(VM_PFNMAP)]      = "pf",
 668                 [ilog2(VM_LOCKED)]      = "lo",
 669                 [ilog2(VM_IO)]          = "io",
 670                 [ilog2(VM_SEQ_READ)]    = "sr",
 671                 [ilog2(VM_RAND_READ)]   = "rr",
 672                 [ilog2(VM_DONTCOPY)]    = "dc",
 673                 [ilog2(VM_DONTEXPAND)]  = "de",
 674                 [ilog2(VM_LOCKONFAULT)] = "lf",
 675                 [ilog2(VM_ACCOUNT)]     = "ac",
 676                 [ilog2(VM_NORESERVE)]   = "nr",
 677                 [ilog2(VM_HUGETLB)]     = "ht",
 678                 [ilog2(VM_SYNC)]        = "sf",
 679                 [ilog2(VM_ARCH_1)]      = "ar",
 680                 [ilog2(VM_WIPEONFORK)]  = "wf",
 681                 [ilog2(VM_DONTDUMP)]    = "dd",
 682 #ifdef CONFIG_ARM64_BTI
 683                 [ilog2(VM_ARM64_BTI)]   = "bt",
 684 #endif
 685 #ifdef CONFIG_MEM_SOFT_DIRTY
 686                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 687 #endif
 688                 [ilog2(VM_MIXEDMAP)]    = "mm",
 689                 [ilog2(VM_HUGEPAGE)]    = "hg",
 690                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 691                 [ilog2(VM_MERGEABLE)]   = "mg",
 692                 [ilog2(VM_UFFD_MISSING)]= "um",
 693                 [ilog2(VM_UFFD_WP)]     = "uw",
 694 #ifdef CONFIG_ARM64_MTE
 695                 [ilog2(VM_MTE)]         = "mt",
 696                 [ilog2(VM_MTE_ALLOWED)] = "",
 697 #endif
 698 #ifdef CONFIG_ARCH_HAS_PKEYS
 699                 /* These come out via ProtectionKey: */
 700                 [ilog2(VM_PKEY_BIT0)]   = "",
 701                 [ilog2(VM_PKEY_BIT1)]   = "",
 702                 [ilog2(VM_PKEY_BIT2)]   = "",
 703                 [ilog2(VM_PKEY_BIT3)]   = "",
 704 #if VM_PKEY_BIT4
 705                 [ilog2(VM_PKEY_BIT4)]   = "",
 706 #endif
 707 #endif /* CONFIG_ARCH_HAS_PKEYS */
 708 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
 709                 [ilog2(VM_UFFD_MINOR)]  = "ui",
 710 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
 711         };
 712         size_t i;
 713
 714         seq_puts(m, "VmFlags: ");
 715         for (i = 0; i < BITS_PER_LONG; i++) {
 716                 if (!mnemonics[i][0])
 717                         continue;
 718                 if (vma->vm_flags & (1UL << i)) {
 719                         seq_putc(m, mnemonics[i][0]);
 720                         seq_putc(m, mnemonics[i][1]);
 721                         seq_putc(m, ' ');
 722                 }
 723         }
 724         seq_putc(m, '\n');
 725 }
 726
 727 #ifdef CONFIG_HUGETLB_PAGE
 728 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 729                                  unsigned long addr, unsigned long end,
 730                                  struct mm_walk *walk)
 731 {
 732         struct mem_size_stats *mss = walk->private;
 733         struct vm_area_struct *vma = walk->vma;
 734         struct page *page = NULL;
 735         pte_t ptent = ptep_get(pte);
 736
 737         if (pte_present(ptent)) {
 738                 page = vm_normal_page(vma, addr, ptent);
 739         } else if (is_swap_pte(ptent)) {
 740                 swp_entry_t swpent = pte_to_swp_entry(ptent);
 741
 742                 if (is_pfn_swap_entry(swpent))
 743                         page = pfn_swap_entry_to_page(swpent);
 744         }
 745         if (page) {
 746                 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
 747                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 748                 else
 749                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 750         }
 751         return 0;
 752 }
 753 #else
 754 #define smaps_hugetlb_range     NULL
 755 #endif /* HUGETLB_PAGE */
 756
 757 static const struct mm_walk_ops smaps_walk_ops = {
 758         .pmd_entry              = smaps_pte_range,
 759         .hugetlb_entry          = smaps_hugetlb_range,
 760         .walk_lock              = PGWALK_RDLOCK,
 761 };
 762
 763 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 764         .pmd_entry              = smaps_pte_range,
 765         .hugetlb_entry          = smaps_hugetlb_range,
 766         .pte_hole               = smaps_pte_hole,
 767         .walk_lock              = PGWALK_RDLOCK,
 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 = ptep_get(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         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1195         if (!pte) {
1196                 walk->action = ACTION_AGAIN;
1197                 return 0;
1198         }
1199         for (; addr != end; pte++, addr += PAGE_SIZE) {
1200                 ptent = ptep_get(pte);
1201
1202                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1203                         clear_soft_dirty(vma, addr, pte);
1204                         continue;
1205                 }
1206
1207                 if (!pte_present(ptent))
1208                         continue;
1209
1210                 page = vm_normal_page(vma, addr, ptent);
1211                 if (!page)
1212                         continue;
1213
1214                 /* Clear accessed and referenced bits. */
1215                 ptep_test_and_clear_young(vma, addr, pte);
1216                 test_and_clear_page_young(page);
1217                 ClearPageReferenced(page);
1218         }
1219         pte_unmap_unlock(pte - 1, ptl);
1220         cond_resched();
1221         return 0;
1222 }
1223
1224 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1225                                 struct mm_walk *walk)
1226 {
1227         struct clear_refs_private *cp = walk->private;
1228         struct vm_area_struct *vma = walk->vma;
1229
1230         if (vma->vm_flags & VM_PFNMAP)
1231                 return 1;
1232
1233         /*
1234          * Writing 1 to /proc/pid/clear_refs affects all pages.
1235          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1236          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1237          * Writing 4 to /proc/pid/clear_refs affects all pages.
1238          */
1239         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1240                 return 1;
1241         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1242                 return 1;
1243         return 0;
1244 }
1245
1246 static const struct mm_walk_ops clear_refs_walk_ops = {
1247         .pmd_entry              = clear_refs_pte_range,
1248         .test_walk              = clear_refs_test_walk,
1249         .walk_lock              = PGWALK_WRLOCK,
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                 VMA_ITERATOR(vmi, mm, 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                         for_each_vma(vmi, vma) {
1301                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1302                                         continue;
1303                                 vm_flags_clear(vma, 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, 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 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1544
1545         /*
1546          * We can assume that @vma always points to a valid one and @end never
1547          * goes beyond vma->vm_end.
1548          */
1549         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1550         if (!pte) {
1551                 walk->action = ACTION_AGAIN;
1552                 return err;
1553         }
1554         for (; addr < end; pte++, addr += PAGE_SIZE) {
1555                 pagemap_entry_t pme;
1556
1557                 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte));
1558                 err = add_to_pagemap(addr, &pme, pm);
1559                 if (err)
1560                         break;
1561         }
1562         pte_unmap_unlock(orig_pte, ptl);
1563
1564         cond_resched();
1565
1566         return err;
1567 }
1568
1569 #ifdef CONFIG_HUGETLB_PAGE
1570 /* This function walks within one hugetlb entry in the single call */
1571 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1572                                  unsigned long addr, unsigned long end,
1573                                  struct mm_walk *walk)
1574 {
1575         struct pagemapread *pm = walk->private;
1576         struct vm_area_struct *vma = walk->vma;
1577         u64 flags = 0, frame = 0;
1578         int err = 0;
1579         pte_t pte;
1580
1581         if (vma->vm_flags & VM_SOFTDIRTY)
1582                 flags |= PM_SOFT_DIRTY;
1583
1584         pte = huge_ptep_get(ptep);
1585         if (pte_present(pte)) {
1586                 struct page *page = pte_page(pte);
1587
1588                 if (!PageAnon(page))
1589                         flags |= PM_FILE;
1590
1591                 if (page_mapcount(page) == 1)
1592                         flags |= PM_MMAP_EXCLUSIVE;
1593
1594                 if (huge_pte_uffd_wp(pte))
1595                         flags |= PM_UFFD_WP;
1596
1597                 flags |= PM_PRESENT;
1598                 if (pm->show_pfn)
1599                         frame = pte_pfn(pte) +
1600                                 ((addr & ~hmask) >> PAGE_SHIFT);
1601         } else if (pte_swp_uffd_wp_any(pte)) {
1602                 flags |= PM_UFFD_WP;
1603         }
1604
1605         for (; addr != end; addr += PAGE_SIZE) {
1606                 pagemap_entry_t pme = make_pme(frame, flags);
1607
1608                 err = add_to_pagemap(addr, &pme, pm);
1609                 if (err)
1610                         return err;
1611                 if (pm->show_pfn && (flags & PM_PRESENT))
1612                         frame++;
1613         }
1614
1615         cond_resched();
1616
1617         return err;
1618 }
1619 #else
1620 #define pagemap_hugetlb_range   NULL
1621 #endif /* HUGETLB_PAGE */
1622
1623 static const struct mm_walk_ops pagemap_ops = {
1624         .pmd_entry      = pagemap_pmd_range,
1625         .pte_hole       = pagemap_pte_hole,
1626         .hugetlb_entry  = pagemap_hugetlb_range,
1627         .walk_lock      = PGWALK_RDLOCK,
1628 };
1629
1630 /*
1631  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1632  *
1633  * For each page in the address space, this file contains one 64-bit entry
1634  * consisting of the following:
1635  *
1636  * Bits 0-54  page frame number (PFN) if present
1637  * Bits 0-4   swap type if swapped
1638  * Bits 5-54  swap offset if swapped
1639  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1640  * Bit  56    page exclusively mapped
1641  * Bit  57    pte is uffd-wp write-protected
1642  * Bits 58-60 zero
1643  * Bit  61    page is file-page or shared-anon
1644  * Bit  62    page swapped
1645  * Bit  63    page present
1646  *
1647  * If the page is not present but in swap, then the PFN contains an
1648  * encoding of the swap file number and the page's offset into the
1649  * swap. Unmapped pages return a null PFN. This allows determining
1650  * precisely which pages are mapped (or in swap) and comparing mapped
1651  * pages between processes.
1652  *
1653  * Efficient users of this interface will use /proc/pid/maps to
1654  * determine which areas of memory are actually mapped and llseek to
1655  * skip over unmapped regions.
1656  */
1657 static ssize_t pagemap_read(struct file *file, char __user *buf,
1658                             size_t count, loff_t *ppos)
1659 {
1660         struct mm_struct *mm = file->private_data;
1661         struct pagemapread pm;
1662         unsigned long src;
1663         unsigned long svpfn;
1664         unsigned long start_vaddr;
1665         unsigned long end_vaddr;
1666         int ret = 0, copied = 0;
1667
1668         if (!mm || !mmget_not_zero(mm))
1669                 goto out;
1670
1671         ret = -EINVAL;
1672         /* file position must be aligned */
1673         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1674                 goto out_mm;
1675
1676         ret = 0;
1677         if (!count)
1678                 goto out_mm;
1679
1680         /* do not disclose physical addresses: attack vector */
1681         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1682
1683         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1684         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1685         ret = -ENOMEM;
1686         if (!pm.buffer)
1687                 goto out_mm;
1688
1689         src = *ppos;
1690         svpfn = src / PM_ENTRY_BYTES;
1691         end_vaddr = mm->task_size;
1692
1693         /* watch out for wraparound */
1694         start_vaddr = end_vaddr;
1695         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
1696                 unsigned long end;
1697
1698                 ret = mmap_read_lock_killable(mm);
1699                 if (ret)
1700                         goto out_free;
1701                 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
1702                 mmap_read_unlock(mm);
1703
1704                 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
1705                 if (end >= start_vaddr && end < mm->task_size)
1706                         end_vaddr = end;
1707         }
1708
1709         /* Ensure the address is inside the task */
1710         if (start_vaddr > mm->task_size)
1711                 start_vaddr = end_vaddr;
1712
1713         ret = 0;
1714         while (count && (start_vaddr < end_vaddr)) {
1715                 int len;
1716                 unsigned long end;
1717
1718                 pm.pos = 0;
1719                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1720                 /* overflow ? */
1721                 if (end < start_vaddr || end > end_vaddr)
1722                         end = end_vaddr;
1723                 ret = mmap_read_lock_killable(mm);
1724                 if (ret)
1725                         goto out_free;
1726                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1727                 mmap_read_unlock(mm);
1728                 start_vaddr = end;
1729
1730                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1731                 if (copy_to_user(buf, pm.buffer, len)) {
1732                         ret = -EFAULT;
1733                         goto out_free;
1734                 }
1735                 copied += len;
1736                 buf += len;
1737                 count -= len;
1738         }
1739         *ppos += copied;
1740         if (!ret || ret == PM_END_OF_BUFFER)
1741                 ret = copied;
1742
1743 out_free:
1744         kfree(pm.buffer);
1745 out_mm:
1746         mmput(mm);
1747 out:
1748         return ret;
1749 }
1750
1751 static int pagemap_open(struct inode *inode, struct file *file)
1752 {
1753         struct mm_struct *mm;
1754
1755         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1756         if (IS_ERR(mm))
1757                 return PTR_ERR(mm);
1758         file->private_data = mm;
1759         return 0;
1760 }
1761
1762 static int pagemap_release(struct inode *inode, struct file *file)
1763 {
1764         struct mm_struct *mm = file->private_data;
1765
1766         if (mm)
1767                 mmdrop(mm);
1768         return 0;
1769 }
1770
1771 const struct file_operations proc_pagemap_operations = {
1772         .llseek         = mem_lseek, /* borrow this */
1773         .read           = pagemap_read,
1774         .open           = pagemap_open,
1775         .release        = pagemap_release,
1776 };
1777 #endif /* CONFIG_PROC_PAGE_MONITOR */
1778
1779 #ifdef CONFIG_NUMA
1780
1781 struct numa_maps {
1782         unsigned long pages;
1783         unsigned long anon;
1784         unsigned long active;
1785         unsigned long writeback;
1786         unsigned long mapcount_max;
1787         unsigned long dirty;
1788         unsigned long swapcache;
1789         unsigned long node[MAX_NUMNODES];
1790 };
1791
1792 struct numa_maps_private {
1793         struct proc_maps_private proc_maps;
1794         struct numa_maps md;
1795 };
1796
1797 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1798                         unsigned long nr_pages)
1799 {
1800         int count = page_mapcount(page);
1801
1802         md->pages += nr_pages;
1803         if (pte_dirty || PageDirty(page))
1804                 md->dirty += nr_pages;
1805
1806         if (PageSwapCache(page))
1807                 md->swapcache += nr_pages;
1808
1809         if (PageActive(page) || PageUnevictable(page))
1810                 md->active += nr_pages;
1811
1812         if (PageWriteback(page))
1813                 md->writeback += nr_pages;
1814
1815         if (PageAnon(page))
1816                 md->anon += nr_pages;
1817
1818         if (count > md->mapcount_max)
1819                 md->mapcount_max = count;
1820
1821         md->node[page_to_nid(page)] += nr_pages;
1822 }
1823
1824 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1825                 unsigned long addr)
1826 {
1827         struct page *page;
1828         int nid;
1829
1830         if (!pte_present(pte))
1831                 return NULL;
1832
1833         page = vm_normal_page(vma, addr, pte);
1834         if (!page || is_zone_device_page(page))
1835                 return NULL;
1836
1837         if (PageReserved(page))
1838                 return NULL;
1839
1840         nid = page_to_nid(page);
1841         if (!node_isset(nid, node_states[N_MEMORY]))
1842                 return NULL;
1843
1844         return page;
1845 }
1846
1847 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1848 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1849                                               struct vm_area_struct *vma,
1850                                               unsigned long addr)
1851 {
1852         struct page *page;
1853         int nid;
1854
1855         if (!pmd_present(pmd))
1856                 return NULL;
1857
1858         page = vm_normal_page_pmd(vma, addr, pmd);
1859         if (!page)
1860                 return NULL;
1861
1862         if (PageReserved(page))
1863                 return NULL;
1864
1865         nid = page_to_nid(page);
1866         if (!node_isset(nid, node_states[N_MEMORY]))
1867                 return NULL;
1868
1869         return page;
1870 }
1871 #endif
1872
1873 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1874                 unsigned long end, struct mm_walk *walk)
1875 {
1876         struct numa_maps *md = walk->private;
1877         struct vm_area_struct *vma = walk->vma;
1878         spinlock_t *ptl;
1879         pte_t *orig_pte;
1880         pte_t *pte;
1881
1882 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1883         ptl = pmd_trans_huge_lock(pmd, vma);
1884         if (ptl) {
1885                 struct page *page;
1886
1887                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1888                 if (page)
1889                         gather_stats(page, md, pmd_dirty(*pmd),
1890                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1891                 spin_unlock(ptl);
1892                 return 0;
1893         }
1894 #endif
1895         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1896         if (!pte) {
1897                 walk->action = ACTION_AGAIN;
1898                 return 0;
1899         }
1900         do {
1901                 pte_t ptent = ptep_get(pte);
1902                 struct page *page = can_gather_numa_stats(ptent, vma, addr);
1903                 if (!page)
1904                         continue;
1905                 gather_stats(page, md, pte_dirty(ptent), 1);
1906
1907         } while (pte++, addr += PAGE_SIZE, addr != end);
1908         pte_unmap_unlock(orig_pte, ptl);
1909         cond_resched();
1910         return 0;
1911 }
1912 #ifdef CONFIG_HUGETLB_PAGE
1913 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1914                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1915 {
1916         pte_t huge_pte = huge_ptep_get(pte);
1917         struct numa_maps *md;
1918         struct page *page;
1919
1920         if (!pte_present(huge_pte))
1921                 return 0;
1922
1923         page = pte_page(huge_pte);
1924
1925         md = walk->private;
1926         gather_stats(page, md, pte_dirty(huge_pte), 1);
1927         return 0;
1928 }
1929
1930 #else
1931 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1932                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1933 {
1934         return 0;
1935 }
1936 #endif
1937
1938 static const struct mm_walk_ops show_numa_ops = {
1939         .hugetlb_entry = gather_hugetlb_stats,
1940         .pmd_entry = gather_pte_stats,
1941         .walk_lock = PGWALK_RDLOCK,
1942 };
1943
1944 /*
1945  * Display pages allocated per node and memory policy via /proc.
1946  */
1947 static int show_numa_map(struct seq_file *m, void *v)
1948 {
1949         struct numa_maps_private *numa_priv = m->private;
1950         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1951         struct vm_area_struct *vma = v;
1952         struct numa_maps *md = &numa_priv->md;
1953         struct file *file = vma->vm_file;
1954         struct mm_struct *mm = vma->vm_mm;
1955         struct mempolicy *pol;
1956         char buffer[64];
1957         int nid;
1958
1959         if (!mm)
1960                 return 0;
1961
1962         /* Ensure we start with an empty set of numa_maps statistics. */
1963         memset(md, 0, sizeof(*md));
1964
1965         pol = __get_vma_policy(vma, vma->vm_start);
1966         if (pol) {
1967                 mpol_to_str(buffer, sizeof(buffer), pol);
1968                 mpol_cond_put(pol);
1969         } else {
1970                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1971         }
1972
1973         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1974
1975         if (file) {
1976                 seq_puts(m, " file=");
1977                 seq_file_path(m, file, "\n\t= ");
1978         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1979                 seq_puts(m, " heap");
1980         } else if (is_stack(vma)) {
1981                 seq_puts(m, " stack");
1982         }
1983
1984         if (is_vm_hugetlb_page(vma))
1985                 seq_puts(m, " huge");
1986
1987         /* mmap_lock is held by m_start */
1988         walk_page_vma(vma, &show_numa_ops, md);
1989
1990         if (!md->pages)
1991                 goto out;
1992
1993         if (md->anon)
1994                 seq_printf(m, " anon=%lu", md->anon);
1995
1996         if (md->dirty)
1997                 seq_printf(m, " dirty=%lu", md->dirty);
1998
1999         if (md->pages != md->anon && md->pages != md->dirty)
2000                 seq_printf(m, " mapped=%lu", md->pages);
2001
2002         if (md->mapcount_max > 1)
2003                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
2004
2005         if (md->swapcache)
2006                 seq_printf(m, " swapcache=%lu", md->swapcache);
2007
2008         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2009                 seq_printf(m, " active=%lu", md->active);
2010
2011         if (md->writeback)
2012                 seq_printf(m, " writeback=%lu", md->writeback);
2013
2014         for_each_node_state(nid, N_MEMORY)
2015                 if (md->node[nid])
2016                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2017
2018         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2019 out:
2020         seq_putc(m, '\n');
2021         return 0;
2022 }
2023
2024 static const struct seq_operations proc_pid_numa_maps_op = {
2025         .start  = m_start,
2026         .next   = m_next,
2027         .stop   = m_stop,
2028         .show   = show_numa_map,
2029 };
2030
2031 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2032 {
2033         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2034                                 sizeof(struct numa_maps_private));
2035 }
2036
2037 const struct file_operations proc_pid_numa_maps_operations = {
2038         .open           = pid_numa_maps_open,
2039         .read           = seq_read,
2040         .llseek         = seq_lseek,
2041         .release        = proc_map_release,
2042 };
2043
2044 #endif /* CONFIG_NUMA */