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_ACCOUNT)]     = "ac",
 678                 [ilog2(VM_NORESERVE)]   = "nr",
 679                 [ilog2(VM_HUGETLB)]     = "ht",
 680                 [ilog2(VM_SYNC)]        = "sf",
 681                 [ilog2(VM_ARCH_1)]      = "ar",
 682                 [ilog2(VM_WIPEONFORK)]  = "wf",
 683                 [ilog2(VM_DONTDUMP)]    = "dd",
 684 #ifdef CONFIG_ARM64_BTI
 685                 [ilog2(VM_ARM64_BTI)]   = "bt",
 686 #endif
 687 #ifdef CONFIG_MEM_SOFT_DIRTY
 688                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 689 #endif
 690                 [ilog2(VM_MIXEDMAP)]    = "mm",
 691                 [ilog2(VM_HUGEPAGE)]    = "hg",
 692                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 693                 [ilog2(VM_MERGEABLE)]   = "mg",
 694                 [ilog2(VM_UFFD_MISSING)]= "um",
 695                 [ilog2(VM_UFFD_WP)]     = "uw",
 696 #ifdef CONFIG_ARM64_MTE
 697                 [ilog2(VM_MTE)]         = "mt",
 698                 [ilog2(VM_MTE_ALLOWED)] = "",
 699 #endif
 700 #ifdef CONFIG_ARCH_HAS_PKEYS
 701                 /* These come out via ProtectionKey: */
 702                 [ilog2(VM_PKEY_BIT0)]   = "",
 703                 [ilog2(VM_PKEY_BIT1)]   = "",
 704                 [ilog2(VM_PKEY_BIT2)]   = "",
 705                 [ilog2(VM_PKEY_BIT3)]   = "",
 706 #if VM_PKEY_BIT4
 707                 [ilog2(VM_PKEY_BIT4)]   = "",
 708 #endif
 709 #endif /* CONFIG_ARCH_HAS_PKEYS */
 710 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
 711                 [ilog2(VM_UFFD_MINOR)]  = "ui",
 712 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
 713         };
 714         size_t i;
 715
 716         seq_puts(m, "VmFlags: ");
 717         for (i = 0; i < BITS_PER_LONG; i++) {
 718                 if (!mnemonics[i][0])
 719                         continue;
 720                 if (vma->vm_flags & (1UL << i)) {
 721                         seq_putc(m, mnemonics[i][0]);
 722                         seq_putc(m, mnemonics[i][1]);
 723                         seq_putc(m, ' ');
 724                 }
 725         }
 726         seq_putc(m, '\n');
 727 }
 728
 729 #ifdef CONFIG_HUGETLB_PAGE
 730 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 731                                  unsigned long addr, unsigned long end,
 732                                  struct mm_walk *walk)
 733 {
 734         struct mem_size_stats *mss = walk->private;
 735         struct vm_area_struct *vma = walk->vma;
 736         struct page *page = NULL;
 737
 738         if (pte_present(*pte)) {
 739                 page = vm_normal_page(vma, addr, *pte);
 740         } else if (is_swap_pte(*pte)) {
 741                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 742
 743                 if (is_pfn_swap_entry(swpent))
 744                         page = pfn_swap_entry_to_page(swpent);
 745         }
 746         if (page) {
 747                 int mapcount = page_mapcount(page);
 748
 749                 if (mapcount >= 2)
 750                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 751                 else
 752                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 753         }
 754         return 0;
 755 }
 756 #else
 757 #define smaps_hugetlb_range     NULL
 758 #endif /* HUGETLB_PAGE */
 759
 760 static const struct mm_walk_ops smaps_walk_ops = {
 761         .pmd_entry              = smaps_pte_range,
 762         .hugetlb_entry          = smaps_hugetlb_range,
 763 };
 764
 765 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 766         .pmd_entry              = smaps_pte_range,
 767         .hugetlb_entry          = smaps_hugetlb_range,
 768         .pte_hole               = smaps_pte_hole,
 769 };
 770
 771 /*
 772  * Gather mem stats from @vma with the indicated beginning
 773  * address @start, and keep them in @mss.
 774  *
 775  * Use vm_start of @vma as the beginning address if @start is 0.
 776  */
 777 static void smap_gather_stats(struct vm_area_struct *vma,
 778                 struct mem_size_stats *mss, unsigned long start)
 779 {
 780         const struct mm_walk_ops *ops = &smaps_walk_ops;
 781
 782         /* Invalid start */
 783         if (start >= vma->vm_end)
 784                 return;
 785
 786 #ifdef CONFIG_SHMEM
 787         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 788                 /*
 789                  * For shared or readonly shmem mappings we know that all
 790                  * swapped out pages belong to the shmem object, and we can
 791                  * obtain the swap value much more efficiently. For private
 792                  * writable mappings, we might have COW pages that are
 793                  * not affected by the parent swapped out pages of the shmem
 794                  * object, so we have to distinguish them during the page walk.
 795                  * Unless we know that the shmem object (or the part mapped by
 796                  * our VMA) has no swapped out pages at all.
 797                  */
 798                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 799
 800                 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 801                                         !(vma->vm_flags & VM_WRITE))) {
 802                         mss->swap += shmem_swapped;
 803                 } else {
 804                         ops = &smaps_shmem_walk_ops;
 805                 }
 806         }
 807 #endif
 808         /* mmap_lock is held in m_start */
 809         if (!start)
 810                 walk_page_vma(vma, ops, mss);
 811         else
 812                 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
 813 }
 814
 815 #define SEQ_PUT_DEC(str, val) \
 816                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 817
 818 /* Show the contents common for smaps and smaps_rollup */
 819 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 820         bool rollup_mode)
 821 {
 822         SEQ_PUT_DEC("Rss:            ", mss->resident);
 823         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 824         SEQ_PUT_DEC(" kB\nPss_Dirty:      ", mss->pss_dirty >> PSS_SHIFT);
 825         if (rollup_mode) {
 826                 /*
 827                  * These are meaningful only for smaps_rollup, otherwise two of
 828                  * them are zero, and the other one is the same as Pss.
 829                  */
 830                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 831                         mss->pss_anon >> PSS_SHIFT);
 832                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 833                         mss->pss_file >> PSS_SHIFT);
 834                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 835                         mss->pss_shmem >> PSS_SHIFT);
 836         }
 837         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 838         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 839         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 840         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 841         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 842         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 843         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 844         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 845         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 846         SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
 847         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 848         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 849                                   mss->private_hugetlb >> 10, 7);
 850         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 851         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 852                                         mss->swap_pss >> PSS_SHIFT);
 853         SEQ_PUT_DEC(" kB\nLocked:         ",
 854                                         mss->pss_locked >> PSS_SHIFT);
 855         seq_puts(m, " kB\n");
 856 }
 857
 858 static int show_smap(struct seq_file *m, void *v)
 859 {
 860         struct vm_area_struct *vma = v;
 861         struct mem_size_stats mss;
 862
 863         memset(&mss, 0, sizeof(mss));
 864
 865         smap_gather_stats(vma, &mss, 0);
 866
 867         show_map_vma(m, vma);
 868
 869         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 870         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 871         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 872         seq_puts(m, " kB\n");
 873
 874         __show_smap(m, &mss, false);
 875
 876         seq_printf(m, "THPeligible:    %d\n",
 877                    hugepage_vma_check(vma, vma->vm_flags, true, false, true));
 878
 879         if (arch_pkeys_enabled())
 880                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 881         show_smap_vma_flags(m, vma);
 882
 883         return 0;
 884 }
 885
 886 static int show_smaps_rollup(struct seq_file *m, void *v)
 887 {
 888         struct proc_maps_private *priv = m->private;
 889         struct mem_size_stats mss;
 890         struct mm_struct *mm = priv->mm;
 891         struct vm_area_struct *vma;
 892         unsigned long vma_start = 0, last_vma_end = 0;
 893         int ret = 0;
 894         MA_STATE(mas, &mm->mm_mt, 0, 0);
 895
 896         priv->task = get_proc_task(priv->inode);
 897         if (!priv->task)
 898                 return -ESRCH;
 899
 900         if (!mm || !mmget_not_zero(mm)) {
 901                 ret = -ESRCH;
 902                 goto out_put_task;
 903         }
 904
 905         memset(&mss, 0, sizeof(mss));
 906
 907         ret = mmap_read_lock_killable(mm);
 908         if (ret)
 909                 goto out_put_mm;
 910
 911         hold_task_mempolicy(priv);
 912         vma = mas_find(&mas, ULONG_MAX);
 913
 914         if (unlikely(!vma))
 915                 goto empty_set;
 916
 917         vma_start = vma->vm_start;
 918         do {
 919                 smap_gather_stats(vma, &mss, 0);
 920                 last_vma_end = vma->vm_end;
 921
 922                 /*
 923                  * Release mmap_lock temporarily if someone wants to
 924                  * access it for write request.
 925                  */
 926                 if (mmap_lock_is_contended(mm)) {
 927                         mas_pause(&mas);
 928                         mmap_read_unlock(mm);
 929                         ret = mmap_read_lock_killable(mm);
 930                         if (ret) {
 931                                 release_task_mempolicy(priv);
 932                                 goto out_put_mm;
 933                         }
 934
 935                         /*
 936                          * After dropping the lock, there are four cases to
 937                          * consider. See the following example for explanation.
 938                          *
 939                          *   +------+------+-----------+
 940                          *   | VMA1 | VMA2 | VMA3      |
 941                          *   +------+------+-----------+
 942                          *   |      |      |           |
 943                          *  4k     8k     16k         400k
 944                          *
 945                          * Suppose we drop the lock after reading VMA2 due to
 946                          * contention, then we get:
 947                          *
 948                          *      last_vma_end = 16k
 949                          *
 950                          * 1) VMA2 is freed, but VMA3 exists:
 951                          *
 952                          *    find_vma(mm, 16k - 1) will return VMA3.
 953                          *    In this case, just continue from VMA3.
 954                          *
 955                          * 2) VMA2 still exists:
 956                          *
 957                          *    find_vma(mm, 16k - 1) will return VMA2.
 958                          *    Iterate the loop like the original one.
 959                          *
 960                          * 3) No more VMAs can be found:
 961                          *
 962                          *    find_vma(mm, 16k - 1) will return NULL.
 963                          *    No more things to do, just break.
 964                          *
 965                          * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
 966                          *
 967                          *    find_vma(mm, 16k - 1) will return VMA' whose range
 968                          *    contains last_vma_end.
 969                          *    Iterate VMA' from last_vma_end.
 970                          */
 971                         vma = mas_find(&mas, ULONG_MAX);
 972                         /* Case 3 above */
 973                         if (!vma)
 974                                 break;
 975
 976                         /* Case 1 above */
 977                         if (vma->vm_start >= last_vma_end)
 978                                 continue;
 979
 980                         /* Case 4 above */
 981                         if (vma->vm_end > last_vma_end)
 982                                 smap_gather_stats(vma, &mss, last_vma_end);
 983                 }
 984                 /* Case 2 above */
 985         } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL);
 986
 987 empty_set:
 988         show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
 989         seq_pad(m, ' ');
 990         seq_puts(m, "[rollup]\n");
 991
 992         __show_smap(m, &mss, true);
 993
 994         release_task_mempolicy(priv);
 995         mmap_read_unlock(mm);
 996
 997 out_put_mm:
 998         mmput(mm);
 999 out_put_task:
1000         put_task_struct(priv->task);
1001         priv->task = NULL;
1002
1003         return ret;
1004 }
1005 #undef SEQ_PUT_DEC
1006
1007 static const struct seq_operations proc_pid_smaps_op = {
1008         .start  = m_start,
1009         .next   = m_next,
1010         .stop   = m_stop,
1011         .show   = show_smap
1012 };
1013
1014 static int pid_smaps_open(struct inode *inode, struct file *file)
1015 {
1016         return do_maps_open(inode, file, &proc_pid_smaps_op);
1017 }
1018
1019 static int smaps_rollup_open(struct inode *inode, struct file *file)
1020 {
1021         int ret;
1022         struct proc_maps_private *priv;
1023
1024         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1025         if (!priv)
1026                 return -ENOMEM;
1027
1028         ret = single_open(file, show_smaps_rollup, priv);
1029         if (ret)
1030                 goto out_free;
1031
1032         priv->inode = inode;
1033         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1034         if (IS_ERR(priv->mm)) {
1035                 ret = PTR_ERR(priv->mm);
1036
1037                 single_release(inode, file);
1038                 goto out_free;
1039         }
1040
1041         return 0;
1042
1043 out_free:
1044         kfree(priv);
1045         return ret;
1046 }
1047
1048 static int smaps_rollup_release(struct inode *inode, struct file *file)
1049 {
1050         struct seq_file *seq = file->private_data;
1051         struct proc_maps_private *priv = seq->private;
1052
1053         if (priv->mm)
1054                 mmdrop(priv->mm);
1055
1056         kfree(priv);
1057         return single_release(inode, file);
1058 }
1059
1060 const struct file_operations proc_pid_smaps_operations = {
1061         .open           = pid_smaps_open,
1062         .read           = seq_read,
1063         .llseek         = seq_lseek,
1064         .release        = proc_map_release,
1065 };
1066
1067 const struct file_operations proc_pid_smaps_rollup_operations = {
1068         .open           = smaps_rollup_open,
1069         .read           = seq_read,
1070         .llseek         = seq_lseek,
1071         .release        = smaps_rollup_release,
1072 };
1073
1074 enum clear_refs_types {
1075         CLEAR_REFS_ALL = 1,
1076         CLEAR_REFS_ANON,
1077         CLEAR_REFS_MAPPED,
1078         CLEAR_REFS_SOFT_DIRTY,
1079         CLEAR_REFS_MM_HIWATER_RSS,
1080         CLEAR_REFS_LAST,
1081 };
1082
1083 struct clear_refs_private {
1084         enum clear_refs_types type;
1085 };
1086
1087 #ifdef CONFIG_MEM_SOFT_DIRTY
1088
1089 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1090 {
1091         struct page *page;
1092
1093         if (!pte_write(pte))
1094                 return false;
1095         if (!is_cow_mapping(vma->vm_flags))
1096                 return false;
1097         if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1098                 return false;
1099         page = vm_normal_page(vma, addr, pte);
1100         if (!page)
1101                 return false;
1102         return page_maybe_dma_pinned(page);
1103 }
1104
1105 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1106                 unsigned long addr, pte_t *pte)
1107 {
1108         /*
1109          * The soft-dirty tracker uses #PF-s to catch writes
1110          * to pages, so write-protect the pte as well. See the
1111          * Documentation/admin-guide/mm/soft-dirty.rst for full description
1112          * of how soft-dirty works.
1113          */
1114         pte_t ptent = *pte;
1115
1116         if (pte_present(ptent)) {
1117                 pte_t old_pte;
1118
1119                 if (pte_is_pinned(vma, addr, ptent))
1120                         return;
1121                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1122                 ptent = pte_wrprotect(old_pte);
1123                 ptent = pte_clear_soft_dirty(ptent);
1124                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1125         } else if (is_swap_pte(ptent)) {
1126                 ptent = pte_swp_clear_soft_dirty(ptent);
1127                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1128         }
1129 }
1130 #else
1131 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1132                 unsigned long addr, pte_t *pte)
1133 {
1134 }
1135 #endif
1136
1137 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1138 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1139                 unsigned long addr, pmd_t *pmdp)
1140 {
1141         pmd_t old, pmd = *pmdp;
1142
1143         if (pmd_present(pmd)) {
1144                 /* See comment in change_huge_pmd() */
1145                 old = pmdp_invalidate(vma, addr, pmdp);
1146                 if (pmd_dirty(old))
1147                         pmd = pmd_mkdirty(pmd);
1148                 if (pmd_young(old))
1149                         pmd = pmd_mkyoung(pmd);
1150
1151                 pmd = pmd_wrprotect(pmd);
1152                 pmd = pmd_clear_soft_dirty(pmd);
1153
1154                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1155         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1156                 pmd = pmd_swp_clear_soft_dirty(pmd);
1157                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1158         }
1159 }
1160 #else
1161 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1162                 unsigned long addr, pmd_t *pmdp)
1163 {
1164 }
1165 #endif
1166
1167 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1168                                 unsigned long end, struct mm_walk *walk)
1169 {
1170         struct clear_refs_private *cp = walk->private;
1171         struct vm_area_struct *vma = walk->vma;
1172         pte_t *pte, ptent;
1173         spinlock_t *ptl;
1174         struct page *page;
1175
1176         ptl = pmd_trans_huge_lock(pmd, vma);
1177         if (ptl) {
1178                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1179                         clear_soft_dirty_pmd(vma, addr, pmd);
1180                         goto out;
1181                 }
1182
1183                 if (!pmd_present(*pmd))
1184                         goto out;
1185
1186                 page = pmd_page(*pmd);
1187
1188                 /* Clear accessed and referenced bits. */
1189                 pmdp_test_and_clear_young(vma, addr, pmd);
1190                 test_and_clear_page_young(page);
1191                 ClearPageReferenced(page);
1192 out:
1193                 spin_unlock(ptl);
1194                 return 0;
1195         }
1196
1197         if (pmd_trans_unstable(pmd))
1198                 return 0;
1199
1200         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1201         for (; addr != end; pte++, addr += PAGE_SIZE) {
1202                 ptent = *pte;
1203
1204                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1205                         clear_soft_dirty(vma, addr, pte);
1206                         continue;
1207                 }
1208
1209                 if (!pte_present(ptent))
1210                         continue;
1211
1212                 page = vm_normal_page(vma, addr, ptent);
1213                 if (!page)
1214                         continue;
1215
1216                 /* Clear accessed and referenced bits. */
1217                 ptep_test_and_clear_young(vma, addr, pte);
1218                 test_and_clear_page_young(page);
1219                 ClearPageReferenced(page);
1220         }
1221         pte_unmap_unlock(pte - 1, ptl);
1222         cond_resched();
1223         return 0;
1224 }
1225
1226 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1227                                 struct mm_walk *walk)
1228 {
1229         struct clear_refs_private *cp = walk->private;
1230         struct vm_area_struct *vma = walk->vma;
1231
1232         if (vma->vm_flags & VM_PFNMAP)
1233                 return 1;
1234
1235         /*
1236          * Writing 1 to /proc/pid/clear_refs affects all pages.
1237          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1238          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1239          * Writing 4 to /proc/pid/clear_refs affects all pages.
1240          */
1241         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1242                 return 1;
1243         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1244                 return 1;
1245         return 0;
1246 }
1247
1248 static const struct mm_walk_ops clear_refs_walk_ops = {
1249         .pmd_entry              = clear_refs_pte_range,
1250         .test_walk              = clear_refs_test_walk,
1251 };
1252
1253 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1254                                 size_t count, loff_t *ppos)
1255 {
1256         struct task_struct *task;
1257         char buffer[PROC_NUMBUF];
1258         struct mm_struct *mm;
1259         struct vm_area_struct *vma;
1260         enum clear_refs_types type;
1261         int itype;
1262         int rv;
1263
1264         memset(buffer, 0, sizeof(buffer));
1265         if (count > sizeof(buffer) - 1)
1266                 count = sizeof(buffer) - 1;
1267         if (copy_from_user(buffer, buf, count))
1268                 return -EFAULT;
1269         rv = kstrtoint(strstrip(buffer), 10, &itype);
1270         if (rv < 0)
1271                 return rv;
1272         type = (enum clear_refs_types)itype;
1273         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1274                 return -EINVAL;
1275
1276         task = get_proc_task(file_inode(file));
1277         if (!task)
1278                 return -ESRCH;
1279         mm = get_task_mm(task);
1280         if (mm) {
1281                 MA_STATE(mas, &mm->mm_mt, 0, 0);
1282                 struct mmu_notifier_range range;
1283                 struct clear_refs_private cp = {
1284                         .type = type,
1285                 };
1286
1287                 if (mmap_write_lock_killable(mm)) {
1288                         count = -EINTR;
1289                         goto out_mm;
1290                 }
1291                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1292                         /*
1293                          * Writing 5 to /proc/pid/clear_refs resets the peak
1294                          * resident set size to this mm's current rss value.
1295                          */
1296                         reset_mm_hiwater_rss(mm);
1297                         goto out_unlock;
1298                 }
1299
1300                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1301                         mas_for_each(&mas, vma, ULONG_MAX) {
1302                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1303                                         continue;
1304                                 vma->vm_flags &= ~VM_SOFTDIRTY;
1305                                 vma_set_page_prot(vma);
1306                         }
1307
1308                         inc_tlb_flush_pending(mm);
1309                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1310                                                 0, NULL, mm, 0, -1UL);
1311                         mmu_notifier_invalidate_range_start(&range);
1312                 }
1313                 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1314                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1315                         mmu_notifier_invalidate_range_end(&range);
1316                         flush_tlb_mm(mm);
1317                         dec_tlb_flush_pending(mm);
1318                 }
1319 out_unlock:
1320                 mmap_write_unlock(mm);
1321 out_mm:
1322                 mmput(mm);
1323         }
1324         put_task_struct(task);
1325
1326         return count;
1327 }
1328
1329 const struct file_operations proc_clear_refs_operations = {
1330         .write          = clear_refs_write,
1331         .llseek         = noop_llseek,
1332 };
1333
1334 typedef struct {
1335         u64 pme;
1336 } pagemap_entry_t;
1337
1338 struct pagemapread {
1339         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1340         pagemap_entry_t *buffer;
1341         bool show_pfn;
1342 };
1343
1344 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1345 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1346
1347 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1348 #define PM_PFRAME_BITS          55
1349 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1350 #define PM_SOFT_DIRTY           BIT_ULL(55)
1351 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1352 #define PM_UFFD_WP              BIT_ULL(57)
1353 #define PM_FILE                 BIT_ULL(61)
1354 #define PM_SWAP                 BIT_ULL(62)
1355 #define PM_PRESENT              BIT_ULL(63)
1356
1357 #define PM_END_OF_BUFFER    1
1358
1359 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1360 {
1361         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1362 }
1363
1364 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1365                           struct pagemapread *pm)
1366 {
1367         pm->buffer[pm->pos++] = *pme;
1368         if (pm->pos >= pm->len)
1369                 return PM_END_OF_BUFFER;
1370         return 0;
1371 }
1372
1373 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1374                             __always_unused int depth, struct mm_walk *walk)
1375 {
1376         struct pagemapread *pm = walk->private;
1377         unsigned long addr = start;
1378         int err = 0;
1379
1380         while (addr < end) {
1381                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1382                 pagemap_entry_t pme = make_pme(0, 0);
1383                 /* End of address space hole, which we mark as non-present. */
1384                 unsigned long hole_end;
1385
1386                 if (vma)
1387                         hole_end = min(end, vma->vm_start);
1388                 else
1389                         hole_end = end;
1390
1391                 for (; addr < hole_end; addr += PAGE_SIZE) {
1392                         err = add_to_pagemap(addr, &pme, pm);
1393                         if (err)
1394                                 goto out;
1395                 }
1396
1397                 if (!vma)
1398                         break;
1399
1400                 /* Addresses in the VMA. */
1401                 if (vma->vm_flags & VM_SOFTDIRTY)
1402                         pme = make_pme(0, PM_SOFT_DIRTY);
1403                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1404                         err = add_to_pagemap(addr, &pme, pm);
1405                         if (err)
1406                                 goto out;
1407                 }
1408         }
1409 out:
1410         return err;
1411 }
1412
1413 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1414                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1415 {
1416         u64 frame = 0, flags = 0;
1417         struct page *page = NULL;
1418         bool migration = false;
1419
1420         if (pte_present(pte)) {
1421                 if (pm->show_pfn)
1422                         frame = pte_pfn(pte);
1423                 flags |= PM_PRESENT;
1424                 page = vm_normal_page(vma, addr, pte);
1425                 if (pte_soft_dirty(pte))
1426                         flags |= PM_SOFT_DIRTY;
1427                 if (pte_uffd_wp(pte))
1428                         flags |= PM_UFFD_WP;
1429         } else if (is_swap_pte(pte)) {
1430                 swp_entry_t entry;
1431                 if (pte_swp_soft_dirty(pte))
1432                         flags |= PM_SOFT_DIRTY;
1433                 if (pte_swp_uffd_wp(pte))
1434                         flags |= PM_UFFD_WP;
1435                 entry = pte_to_swp_entry(pte);
1436                 if (pm->show_pfn) {
1437                         pgoff_t offset;
1438                         /*
1439                          * For PFN swap offsets, keeping the offset field
1440                          * to be PFN only to be compatible with old smaps.
1441                          */
1442                         if (is_pfn_swap_entry(entry))
1443                                 offset = swp_offset_pfn(entry);
1444                         else
1445                                 offset = swp_offset(entry);
1446                         frame = swp_type(entry) |
1447                             (offset << MAX_SWAPFILES_SHIFT);
1448                 }
1449                 flags |= PM_SWAP;
1450                 migration = is_migration_entry(entry);
1451                 if (is_pfn_swap_entry(entry))
1452                         page = pfn_swap_entry_to_page(entry);
1453                 if (pte_marker_entry_uffd_wp(entry))
1454                         flags |= PM_UFFD_WP;
1455         }
1456
1457         if (page && !PageAnon(page))
1458                 flags |= PM_FILE;
1459         if (page && !migration && page_mapcount(page) == 1)
1460                 flags |= PM_MMAP_EXCLUSIVE;
1461         if (vma->vm_flags & VM_SOFTDIRTY)
1462                 flags |= PM_SOFT_DIRTY;
1463
1464         return make_pme(frame, flags);
1465 }
1466
1467 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1468                              struct mm_walk *walk)
1469 {
1470         struct vm_area_struct *vma = walk->vma;
1471         struct pagemapread *pm = walk->private;
1472         spinlock_t *ptl;
1473         pte_t *pte, *orig_pte;
1474         int err = 0;
1475 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1476         bool migration = false;
1477
1478         ptl = pmd_trans_huge_lock(pmdp, vma);
1479         if (ptl) {
1480                 u64 flags = 0, frame = 0;
1481                 pmd_t pmd = *pmdp;
1482                 struct page *page = NULL;
1483
1484                 if (vma->vm_flags & VM_SOFTDIRTY)
1485                         flags |= PM_SOFT_DIRTY;
1486
1487                 if (pmd_present(pmd)) {
1488                         page = pmd_page(pmd);
1489
1490                         flags |= PM_PRESENT;
1491                         if (pmd_soft_dirty(pmd))
1492                                 flags |= PM_SOFT_DIRTY;
1493                         if (pmd_uffd_wp(pmd))
1494                                 flags |= PM_UFFD_WP;
1495                         if (pm->show_pfn)
1496                                 frame = pmd_pfn(pmd) +
1497                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1498                 }
1499 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1500                 else if (is_swap_pmd(pmd)) {
1501                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1502                         unsigned long offset;
1503
1504                         if (pm->show_pfn) {
1505                                 if (is_pfn_swap_entry(entry))
1506                                         offset = swp_offset_pfn(entry);
1507                                 else
1508                                         offset = swp_offset(entry);
1509                                 offset = offset +
1510                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1511                                 frame = swp_type(entry) |
1512                                         (offset << MAX_SWAPFILES_SHIFT);
1513                         }
1514                         flags |= PM_SWAP;
1515                         if (pmd_swp_soft_dirty(pmd))
1516                                 flags |= PM_SOFT_DIRTY;
1517                         if (pmd_swp_uffd_wp(pmd))
1518                                 flags |= PM_UFFD_WP;
1519                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1520                         migration = is_migration_entry(entry);
1521                         page = pfn_swap_entry_to_page(entry);
1522                 }
1523 #endif
1524
1525                 if (page && !migration && page_mapcount(page) == 1)
1526                         flags |= PM_MMAP_EXCLUSIVE;
1527
1528                 for (; addr != end; addr += PAGE_SIZE) {
1529                         pagemap_entry_t pme = make_pme(frame, flags);
1530
1531                         err = add_to_pagemap(addr, &pme, pm);
1532                         if (err)
1533                                 break;
1534                         if (pm->show_pfn) {
1535                                 if (flags & PM_PRESENT)
1536                                         frame++;
1537                                 else if (flags & PM_SWAP)
1538                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1539                         }
1540                 }
1541                 spin_unlock(ptl);
1542                 return err;
1543         }
1544
1545         if (pmd_trans_unstable(pmdp))
1546                 return 0;
1547 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1548
1549         /*
1550          * We can assume that @vma always points to a valid one and @end never
1551          * goes beyond vma->vm_end.
1552          */
1553         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1554         for (; addr < end; pte++, addr += PAGE_SIZE) {
1555                 pagemap_entry_t pme;
1556
1557                 pme = pte_to_pagemap_entry(pm, vma, addr, *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 };
1628
1629 /*
1630  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1631  *
1632  * For each page in the address space, this file contains one 64-bit entry
1633  * consisting of the following:
1634  *
1635  * Bits 0-54  page frame number (PFN) if present
1636  * Bits 0-4   swap type if swapped
1637  * Bits 5-54  swap offset if swapped
1638  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1639  * Bit  56    page exclusively mapped
1640  * Bit  57    pte is uffd-wp write-protected
1641  * Bits 58-60 zero
1642  * Bit  61    page is file-page or shared-anon
1643  * Bit  62    page swapped
1644  * Bit  63    page present
1645  *
1646  * If the page is not present but in swap, then the PFN contains an
1647  * encoding of the swap file number and the page's offset into the
1648  * swap. Unmapped pages return a null PFN. This allows determining
1649  * precisely which pages are mapped (or in swap) and comparing mapped
1650  * pages between processes.
1651  *
1652  * Efficient users of this interface will use /proc/pid/maps to
1653  * determine which areas of memory are actually mapped and llseek to
1654  * skip over unmapped regions.
1655  */
1656 static ssize_t pagemap_read(struct file *file, char __user *buf,
1657                             size_t count, loff_t *ppos)
1658 {
1659         struct mm_struct *mm = file->private_data;
1660         struct pagemapread pm;
1661         unsigned long src;
1662         unsigned long svpfn;
1663         unsigned long start_vaddr;
1664         unsigned long end_vaddr;
1665         int ret = 0, copied = 0;
1666
1667         if (!mm || !mmget_not_zero(mm))
1668                 goto out;
1669
1670         ret = -EINVAL;
1671         /* file position must be aligned */
1672         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1673                 goto out_mm;
1674
1675         ret = 0;
1676         if (!count)
1677                 goto out_mm;
1678
1679         /* do not disclose physical addresses: attack vector */
1680         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1681
1682         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1683         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1684         ret = -ENOMEM;
1685         if (!pm.buffer)
1686                 goto out_mm;
1687
1688         src = *ppos;
1689         svpfn = src / PM_ENTRY_BYTES;
1690         end_vaddr = mm->task_size;
1691
1692         /* watch out for wraparound */
1693         start_vaddr = end_vaddr;
1694         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT))
1695                 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
1696
1697         /* Ensure the address is inside the task */
1698         if (start_vaddr > mm->task_size)
1699                 start_vaddr = end_vaddr;
1700
1701         /*
1702          * The odds are that this will stop walking way
1703          * before end_vaddr, because the length of the
1704          * user buffer is tracked in "pm", and the walk
1705          * will stop when we hit the end of the buffer.
1706          */
1707         ret = 0;
1708         while (count && (start_vaddr < end_vaddr)) {
1709                 int len;
1710                 unsigned long end;
1711
1712                 pm.pos = 0;
1713                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1714                 /* overflow ? */
1715                 if (end < start_vaddr || end > end_vaddr)
1716                         end = end_vaddr;
1717                 ret = mmap_read_lock_killable(mm);
1718                 if (ret)
1719                         goto out_free;
1720                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1721                 mmap_read_unlock(mm);
1722                 start_vaddr = end;
1723
1724                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1725                 if (copy_to_user(buf, pm.buffer, len)) {
1726                         ret = -EFAULT;
1727                         goto out_free;
1728                 }
1729                 copied += len;
1730                 buf += len;
1731                 count -= len;
1732         }
1733         *ppos += copied;
1734         if (!ret || ret == PM_END_OF_BUFFER)
1735                 ret = copied;
1736
1737 out_free:
1738         kfree(pm.buffer);
1739 out_mm:
1740         mmput(mm);
1741 out:
1742         return ret;
1743 }
1744
1745 static int pagemap_open(struct inode *inode, struct file *file)
1746 {
1747         struct mm_struct *mm;
1748
1749         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1750         if (IS_ERR(mm))
1751                 return PTR_ERR(mm);
1752         file->private_data = mm;
1753         return 0;
1754 }
1755
1756 static int pagemap_release(struct inode *inode, struct file *file)
1757 {
1758         struct mm_struct *mm = file->private_data;
1759
1760         if (mm)
1761                 mmdrop(mm);
1762         return 0;
1763 }
1764
1765 const struct file_operations proc_pagemap_operations = {
1766         .llseek         = mem_lseek, /* borrow this */
1767         .read           = pagemap_read,
1768         .open           = pagemap_open,
1769         .release        = pagemap_release,
1770 };
1771 #endif /* CONFIG_PROC_PAGE_MONITOR */
1772
1773 #ifdef CONFIG_NUMA
1774
1775 struct numa_maps {
1776         unsigned long pages;
1777         unsigned long anon;
1778         unsigned long active;
1779         unsigned long writeback;
1780         unsigned long mapcount_max;
1781         unsigned long dirty;
1782         unsigned long swapcache;
1783         unsigned long node[MAX_NUMNODES];
1784 };
1785
1786 struct numa_maps_private {
1787         struct proc_maps_private proc_maps;
1788         struct numa_maps md;
1789 };
1790
1791 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1792                         unsigned long nr_pages)
1793 {
1794         int count = page_mapcount(page);
1795
1796         md->pages += nr_pages;
1797         if (pte_dirty || PageDirty(page))
1798                 md->dirty += nr_pages;
1799
1800         if (PageSwapCache(page))
1801                 md->swapcache += nr_pages;
1802
1803         if (PageActive(page) || PageUnevictable(page))
1804                 md->active += nr_pages;
1805
1806         if (PageWriteback(page))
1807                 md->writeback += nr_pages;
1808
1809         if (PageAnon(page))
1810                 md->anon += nr_pages;
1811
1812         if (count > md->mapcount_max)
1813                 md->mapcount_max = count;
1814
1815         md->node[page_to_nid(page)] += nr_pages;
1816 }
1817
1818 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1819                 unsigned long addr)
1820 {
1821         struct page *page;
1822         int nid;
1823
1824         if (!pte_present(pte))
1825                 return NULL;
1826
1827         page = vm_normal_page(vma, addr, pte);
1828         if (!page || is_zone_device_page(page))
1829                 return NULL;
1830
1831         if (PageReserved(page))
1832                 return NULL;
1833
1834         nid = page_to_nid(page);
1835         if (!node_isset(nid, node_states[N_MEMORY]))
1836                 return NULL;
1837
1838         return page;
1839 }
1840
1841 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1842 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1843                                               struct vm_area_struct *vma,
1844                                               unsigned long addr)
1845 {
1846         struct page *page;
1847         int nid;
1848
1849         if (!pmd_present(pmd))
1850                 return NULL;
1851
1852         page = vm_normal_page_pmd(vma, addr, pmd);
1853         if (!page)
1854                 return NULL;
1855
1856         if (PageReserved(page))
1857                 return NULL;
1858
1859         nid = page_to_nid(page);
1860         if (!node_isset(nid, node_states[N_MEMORY]))
1861                 return NULL;
1862
1863         return page;
1864 }
1865 #endif
1866
1867 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1868                 unsigned long end, struct mm_walk *walk)
1869 {
1870         struct numa_maps *md = walk->private;
1871         struct vm_area_struct *vma = walk->vma;
1872         spinlock_t *ptl;
1873         pte_t *orig_pte;
1874         pte_t *pte;
1875
1876 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1877         ptl = pmd_trans_huge_lock(pmd, vma);
1878         if (ptl) {
1879                 struct page *page;
1880
1881                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1882                 if (page)
1883                         gather_stats(page, md, pmd_dirty(*pmd),
1884                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1885                 spin_unlock(ptl);
1886                 return 0;
1887         }
1888
1889         if (pmd_trans_unstable(pmd))
1890                 return 0;
1891 #endif
1892         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1893         do {
1894                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1895                 if (!page)
1896                         continue;
1897                 gather_stats(page, md, pte_dirty(*pte), 1);
1898
1899         } while (pte++, addr += PAGE_SIZE, addr != end);
1900         pte_unmap_unlock(orig_pte, ptl);
1901         cond_resched();
1902         return 0;
1903 }
1904 #ifdef CONFIG_HUGETLB_PAGE
1905 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1906                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1907 {
1908         pte_t huge_pte = huge_ptep_get(pte);
1909         struct numa_maps *md;
1910         struct page *page;
1911
1912         if (!pte_present(huge_pte))
1913                 return 0;
1914
1915         page = pte_page(huge_pte);
1916
1917         md = walk->private;
1918         gather_stats(page, md, pte_dirty(huge_pte), 1);
1919         return 0;
1920 }
1921
1922 #else
1923 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1924                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1925 {
1926         return 0;
1927 }
1928 #endif
1929
1930 static const struct mm_walk_ops show_numa_ops = {
1931         .hugetlb_entry = gather_hugetlb_stats,
1932         .pmd_entry = gather_pte_stats,
1933 };
1934
1935 /*
1936  * Display pages allocated per node and memory policy via /proc.
1937  */
1938 static int show_numa_map(struct seq_file *m, void *v)
1939 {
1940         struct numa_maps_private *numa_priv = m->private;
1941         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1942         struct vm_area_struct *vma = v;
1943         struct numa_maps *md = &numa_priv->md;
1944         struct file *file = vma->vm_file;
1945         struct mm_struct *mm = vma->vm_mm;
1946         struct mempolicy *pol;
1947         char buffer[64];
1948         int nid;
1949
1950         if (!mm)
1951                 return 0;
1952
1953         /* Ensure we start with an empty set of numa_maps statistics. */
1954         memset(md, 0, sizeof(*md));
1955
1956         pol = __get_vma_policy(vma, vma->vm_start);
1957         if (pol) {
1958                 mpol_to_str(buffer, sizeof(buffer), pol);
1959                 mpol_cond_put(pol);
1960         } else {
1961                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1962         }
1963
1964         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1965
1966         if (file) {
1967                 seq_puts(m, " file=");
1968                 seq_file_path(m, file, "\n\t= ");
1969         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1970                 seq_puts(m, " heap");
1971         } else if (is_stack(vma)) {
1972                 seq_puts(m, " stack");
1973         }
1974
1975         if (is_vm_hugetlb_page(vma))
1976                 seq_puts(m, " huge");
1977
1978         /* mmap_lock is held by m_start */
1979         walk_page_vma(vma, &show_numa_ops, md);
1980
1981         if (!md->pages)
1982                 goto out;
1983
1984         if (md->anon)
1985                 seq_printf(m, " anon=%lu", md->anon);
1986
1987         if (md->dirty)
1988                 seq_printf(m, " dirty=%lu", md->dirty);
1989
1990         if (md->pages != md->anon && md->pages != md->dirty)
1991                 seq_printf(m, " mapped=%lu", md->pages);
1992
1993         if (md->mapcount_max > 1)
1994                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1995
1996         if (md->swapcache)
1997                 seq_printf(m, " swapcache=%lu", md->swapcache);
1998
1999         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2000                 seq_printf(m, " active=%lu", md->active);
2001
2002         if (md->writeback)
2003                 seq_printf(m, " writeback=%lu", md->writeback);
2004
2005         for_each_node_state(nid, N_MEMORY)
2006                 if (md->node[nid])
2007                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2008
2009         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2010 out:
2011         seq_putc(m, '\n');
2012         return 0;
2013 }
2014
2015 static const struct seq_operations proc_pid_numa_maps_op = {
2016         .start  = m_start,
2017         .next   = m_next,
2018         .stop   = m_stop,
2019         .show   = show_numa_map,
2020 };
2021
2022 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2023 {
2024         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2025                                 sizeof(struct numa_maps_private));
2026 }
2027
2028 const struct file_operations proc_pid_numa_maps_operations = {
2029         .open           = pid_numa_maps_open,
2030         .read           = seq_read,
2031         .llseek         = seq_lseek,
2032         .release        = proc_map_release,
2033 };
2034
2035 #endif /* CONFIG_NUMA */