fs/proc/task_mmu.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/pagewalk.h>
   3 #include <linux/mm_inline.h>
   4 #include <linux/hugetlb.h>
   5 #include <linux/huge_mm.h>
   6 #include <linux/mount.h>
   7 #include <linux/ksm.h>
   8 #include <linux/seq_file.h>
   9 #include <linux/highmem.h>
  10 #include <linux/ptrace.h>
  11 #include <linux/slab.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/mempolicy.h>
  14 #include <linux/rmap.h>
  15 #include <linux/swap.h>
  16 #include <linux/sched/mm.h>
  17 #include <linux/swapops.h>
  18 #include <linux/mmu_notifier.h>
  19 #include <linux/page_idle.h>
  20 #include <linux/shmem_fs.h>
  21 #include <linux/uaccess.h>
  22 #include <linux/pkeys.h>
  23 #include <linux/minmax.h>
  24 #include <linux/overflow.h>
  25
  26 #include <asm/elf.h>
  27 #include <asm/tlb.h>
  28 #include <asm/tlbflush.h>
  29 #include "internal.h"
  30
  31 #define SEQ_PUT_DEC(str, val) \
  32                 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
  33 void task_mem(struct seq_file *m, struct mm_struct *mm)
  34 {
  35         unsigned long text, lib, swap, anon, file, shmem;
  36         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  37
  38         anon = get_mm_counter(mm, MM_ANONPAGES);
  39         file = get_mm_counter(mm, MM_FILEPAGES);
  40         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  41
  42         /*
  43          * Note: to minimize their overhead, mm maintains hiwater_vm and
  44          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  45          * collector of these hiwater stats must therefore get total_vm
  46          * and rss too, which will usually be the higher.  Barriers? not
  47          * worth the effort, such snapshots can always be inconsistent.
  48          */
  49         hiwater_vm = total_vm = mm->total_vm;
  50         if (hiwater_vm < mm->hiwater_vm)
  51                 hiwater_vm = mm->hiwater_vm;
  52         hiwater_rss = total_rss = anon + file + shmem;
  53         if (hiwater_rss < mm->hiwater_rss)
  54                 hiwater_rss = mm->hiwater_rss;
  55
  56         /* split executable areas between text and lib */
  57         text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
  58         text = min(text, mm->exec_vm << PAGE_SHIFT);
  59         lib = (mm->exec_vm << PAGE_SHIFT) - text;
  60
  61         swap = get_mm_counter(mm, MM_SWAPENTS);
  62         SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
  63         SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
  64         SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
  65         SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
  66         SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
  67         SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
  68         SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
  69         SEQ_PUT_DEC(" kB\nRssFile:\t", file);
  70         SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
  71         SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
  72         SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
  73         seq_put_decimal_ull_width(m,
  74                     " kB\nVmExe:\t", text >> 10, 8);
  75         seq_put_decimal_ull_width(m,
  76                     " kB\nVmLib:\t", lib >> 10, 8);
  77         seq_put_decimal_ull_width(m,
  78                     " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
  79         SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
  80         seq_puts(m, " kB\n");
  81         hugetlb_report_usage(m, mm);
  82 }
  83 #undef SEQ_PUT_DEC
  84
  85 unsigned long task_vsize(struct mm_struct *mm)
  86 {
  87         return PAGE_SIZE * mm->total_vm;
  88 }
  89
  90 unsigned long task_statm(struct mm_struct *mm,
  91                          unsigned long *shared, unsigned long *text,
  92                          unsigned long *data, unsigned long *resident)
  93 {
  94         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  95                         get_mm_counter(mm, MM_SHMEMPAGES);
  96         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  97                                                                 >> PAGE_SHIFT;
  98         *data = mm->data_vm + mm->stack_vm;
  99         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
 100         return mm->total_vm;
 101 }
 102
 103 #ifdef CONFIG_NUMA
 104 /*
 105  * Save get_task_policy() for show_numa_map().
 106  */
 107 static void hold_task_mempolicy(struct proc_maps_private *priv)
 108 {
 109         struct task_struct *task = priv->task;
 110
 111         task_lock(task);
 112         priv->task_mempolicy = get_task_policy(task);
 113         mpol_get(priv->task_mempolicy);
 114         task_unlock(task);
 115 }
 116 static void release_task_mempolicy(struct proc_maps_private *priv)
 117 {
 118         mpol_put(priv->task_mempolicy);
 119 }
 120 #else
 121 static void hold_task_mempolicy(struct proc_maps_private *priv)
 122 {
 123 }
 124 static void release_task_mempolicy(struct proc_maps_private *priv)
 125 {
 126 }
 127 #endif
 128
 129 static struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv,
 130                                                 loff_t *ppos)
 131 {
 132         struct vm_area_struct *vma = vma_next(&priv->iter);
 133
 134         if (vma) {
 135                 *ppos = vma->vm_start;
 136         } else {
 137                 *ppos = -2UL;
 138                 vma = get_gate_vma(priv->mm);
 139         }
 140
 141         return vma;
 142 }
 143
 144 static void *m_start(struct seq_file *m, loff_t *ppos)
 145 {
 146         struct proc_maps_private *priv = m->private;
 147         unsigned long last_addr = *ppos;
 148         struct mm_struct *mm;
 149
 150         /* See m_next(). Zero at the start or after lseek. */
 151         if (last_addr == -1UL)
 152                 return NULL;
 153
 154         priv->task = get_proc_task(priv->inode);
 155         if (!priv->task)
 156                 return ERR_PTR(-ESRCH);
 157
 158         mm = priv->mm;
 159         if (!mm || !mmget_not_zero(mm)) {
 160                 put_task_struct(priv->task);
 161                 priv->task = NULL;
 162                 return NULL;
 163         }
 164
 165         if (mmap_read_lock_killable(mm)) {
 166                 mmput(mm);
 167                 put_task_struct(priv->task);
 168                 priv->task = NULL;
 169                 return ERR_PTR(-EINTR);
 170         }
 171
 172         vma_iter_init(&priv->iter, mm, last_addr);
 173         hold_task_mempolicy(priv);
 174         if (last_addr == -2UL)
 175                 return get_gate_vma(mm);
 176
 177         return proc_get_vma(priv, ppos);
 178 }
 179
 180 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
 181 {
 182         if (*ppos == -2UL) {
 183                 *ppos = -1UL;
 184                 return NULL;
 185         }
 186         return proc_get_vma(m->private, ppos);
 187 }
 188
 189 static void m_stop(struct seq_file *m, void *v)
 190 {
 191         struct proc_maps_private *priv = m->private;
 192         struct mm_struct *mm = priv->mm;
 193
 194         if (!priv->task)
 195                 return;
 196
 197         release_task_mempolicy(priv);
 198         mmap_read_unlock(mm);
 199         mmput(mm);
 200         put_task_struct(priv->task);
 201         priv->task = NULL;
 202 }
 203
 204 static int proc_maps_open(struct inode *inode, struct file *file,
 205                         const struct seq_operations *ops, int psize)
 206 {
 207         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 208
 209         if (!priv)
 210                 return -ENOMEM;
 211
 212         priv->inode = inode;
 213         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 214         if (IS_ERR(priv->mm)) {
 215                 int err = PTR_ERR(priv->mm);
 216
 217                 seq_release_private(inode, file);
 218                 return err;
 219         }
 220
 221         return 0;
 222 }
 223
 224 static int proc_map_release(struct inode *inode, struct file *file)
 225 {
 226         struct seq_file *seq = file->private_data;
 227         struct proc_maps_private *priv = seq->private;
 228
 229         if (priv->mm)
 230                 mmdrop(priv->mm);
 231
 232         return seq_release_private(inode, file);
 233 }
 234
 235 static int do_maps_open(struct inode *inode, struct file *file,
 236                         const struct seq_operations *ops)
 237 {
 238         return proc_maps_open(inode, file, ops,
 239                                 sizeof(struct proc_maps_private));
 240 }
 241
 242 static void show_vma_header_prefix(struct seq_file *m,
 243                                    unsigned long start, unsigned long end,
 244                                    vm_flags_t flags, unsigned long long pgoff,
 245                                    dev_t dev, unsigned long ino)
 246 {
 247         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 248         seq_put_hex_ll(m, NULL, start, 8);
 249         seq_put_hex_ll(m, "-", end, 8);
 250         seq_putc(m, ' ');
 251         seq_putc(m, flags & VM_READ ? 'r' : '-');
 252         seq_putc(m, flags & VM_WRITE ? 'w' : '-');
 253         seq_putc(m, flags & VM_EXEC ? 'x' : '-');
 254         seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
 255         seq_put_hex_ll(m, " ", pgoff, 8);
 256         seq_put_hex_ll(m, " ", MAJOR(dev), 2);
 257         seq_put_hex_ll(m, ":", MINOR(dev), 2);
 258         seq_put_decimal_ull(m, " ", ino);
 259         seq_putc(m, ' ');
 260 }
 261
 262 static void
 263 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 264 {
 265         struct anon_vma_name *anon_name = NULL;
 266         struct mm_struct *mm = vma->vm_mm;
 267         struct file *file = vma->vm_file;
 268         vm_flags_t flags = vma->vm_flags;
 269         unsigned long ino = 0;
 270         unsigned long long pgoff = 0;
 271         unsigned long start, end;
 272         dev_t dev = 0;
 273         const char *name = NULL;
 274
 275         if (file) {
 276                 struct inode *inode = file_inode(vma->vm_file);
 277                 dev = inode->i_sb->s_dev;
 278                 ino = inode->i_ino;
 279                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 280         }
 281
 282         start = vma->vm_start;
 283         end = vma->vm_end;
 284         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 285         if (mm)
 286                 anon_name = anon_vma_name(vma);
 287
 288         /*
 289          * Print the dentry name for named mappings, and a
 290          * special [heap] marker for the heap:
 291          */
 292         if (file) {
 293                 seq_pad(m, ' ');
 294                 /*
 295                  * If user named this anon shared memory via
 296                  * prctl(PR_SET_VMA ..., use the provided name.
 297                  */
 298                 if (anon_name)
 299                         seq_printf(m, "[anon_shmem:%s]", anon_name->name);
 300                 else
 301                         seq_path(m, file_user_path(file), "\n");
 302                 goto done;
 303         }
 304
 305         if (vma->vm_ops && vma->vm_ops->name) {
 306                 name = vma->vm_ops->name(vma);
 307                 if (name)
 308                         goto done;
 309         }
 310
 311         name = arch_vma_name(vma);
 312         if (!name) {
 313                 if (!mm) {
 314                         name = "[vdso]";
 315                         goto done;
 316                 }
 317
 318                 if (vma_is_initial_heap(vma)) {
 319                         name = "[heap]";
 320                         goto done;
 321                 }
 322
 323                 if (vma_is_initial_stack(vma)) {
 324                         name = "[stack]";
 325                         goto done;
 326                 }
 327
 328                 if (anon_name) {
 329                         seq_pad(m, ' ');
 330                         seq_printf(m, "[anon:%s]", anon_name->name);
 331                 }
 332         }
 333
 334 done:
 335         if (name) {
 336                 seq_pad(m, ' ');
 337                 seq_puts(m, name);
 338         }
 339         seq_putc(m, '\n');
 340 }
 341
 342 static int show_map(struct seq_file *m, void *v)
 343 {
 344         show_map_vma(m, v);
 345         return 0;
 346 }
 347
 348 static const struct seq_operations proc_pid_maps_op = {
 349         .start  = m_start,
 350         .next   = m_next,
 351         .stop   = m_stop,
 352         .show   = show_map
 353 };
 354
 355 static int pid_maps_open(struct inode *inode, struct file *file)
 356 {
 357         return do_maps_open(inode, file, &proc_pid_maps_op);
 358 }
 359
 360 const struct file_operations proc_pid_maps_operations = {
 361         .open           = pid_maps_open,
 362         .read           = seq_read,
 363         .llseek         = seq_lseek,
 364         .release        = proc_map_release,
 365 };
 366
 367 /*
 368  * Proportional Set Size(PSS): my share of RSS.
 369  *
 370  * PSS of a process is the count of pages it has in memory, where each
 371  * page is divided by the number of processes sharing it.  So if a
 372  * process has 1000 pages all to itself, and 1000 shared with one other
 373  * process, its PSS will be 1500.
 374  *
 375  * To keep (accumulated) division errors low, we adopt a 64bit
 376  * fixed-point pss counter to minimize division errors. So (pss >>
 377  * PSS_SHIFT) would be the real byte count.
 378  *
 379  * A shift of 12 before division means (assuming 4K page size):
 380  *      - 1M 3-user-pages add up to 8KB errors;
 381  *      - supports mapcount up to 2^24, or 16M;
 382  *      - supports PSS up to 2^52 bytes, or 4PB.
 383  */
 384 #define PSS_SHIFT 12
 385
 386 #ifdef CONFIG_PROC_PAGE_MONITOR
 387 struct mem_size_stats {
 388         unsigned long resident;
 389         unsigned long shared_clean;
 390         unsigned long shared_dirty;
 391         unsigned long private_clean;
 392         unsigned long private_dirty;
 393         unsigned long referenced;
 394         unsigned long anonymous;
 395         unsigned long lazyfree;
 396         unsigned long anonymous_thp;
 397         unsigned long shmem_thp;
 398         unsigned long file_thp;
 399         unsigned long swap;
 400         unsigned long shared_hugetlb;
 401         unsigned long private_hugetlb;
 402         unsigned long ksm;
 403         u64 pss;
 404         u64 pss_anon;
 405         u64 pss_file;
 406         u64 pss_shmem;
 407         u64 pss_dirty;
 408         u64 pss_locked;
 409         u64 swap_pss;
 410 };
 411
 412 static void smaps_page_accumulate(struct mem_size_stats *mss,
 413                 struct page *page, unsigned long size, unsigned long pss,
 414                 bool dirty, bool locked, bool private)
 415 {
 416         mss->pss += pss;
 417
 418         if (PageAnon(page))
 419                 mss->pss_anon += pss;
 420         else if (PageSwapBacked(page))
 421                 mss->pss_shmem += pss;
 422         else
 423                 mss->pss_file += pss;
 424
 425         if (locked)
 426                 mss->pss_locked += pss;
 427
 428         if (dirty || PageDirty(page)) {
 429                 mss->pss_dirty += pss;
 430                 if (private)
 431                         mss->private_dirty += size;
 432                 else
 433                         mss->shared_dirty += size;
 434         } else {
 435                 if (private)
 436                         mss->private_clean += size;
 437                 else
 438                         mss->shared_clean += size;
 439         }
 440 }
 441
 442 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 443                 bool compound, bool young, bool dirty, bool locked,
 444                 bool migration)
 445 {
 446         int i, nr = compound ? compound_nr(page) : 1;
 447         unsigned long size = nr * PAGE_SIZE;
 448
 449         /*
 450          * First accumulate quantities that depend only on |size| and the type
 451          * of the compound page.
 452          */
 453         if (PageAnon(page)) {
 454                 mss->anonymous += size;
 455                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 456                         mss->lazyfree += size;
 457         }
 458
 459         if (PageKsm(page))
 460                 mss->ksm += size;
 461
 462         mss->resident += size;
 463         /* Accumulate the size in pages that have been accessed. */
 464         if (young || page_is_young(page) || PageReferenced(page))
 465                 mss->referenced += size;
 466
 467         /*
 468          * Then accumulate quantities that may depend on sharing, or that may
 469          * differ page-by-page.
 470          *
 471          * page_count(page) == 1 guarantees the page is mapped exactly once.
 472          * If any subpage of the compound page mapped with PTE it would elevate
 473          * page_count().
 474          *
 475          * The page_mapcount() is called to get a snapshot of the mapcount.
 476          * Without holding the page lock this snapshot can be slightly wrong as
 477          * we cannot always read the mapcount atomically.  It is not safe to
 478          * call page_mapcount() even with PTL held if the page is not mapped,
 479          * especially for migration entries.  Treat regular migration entries
 480          * as mapcount == 1.
 481          */
 482         if ((page_count(page) == 1) || migration) {
 483                 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
 484                         locked, true);
 485                 return;
 486         }
 487         for (i = 0; i < nr; i++, page++) {
 488                 int mapcount = page_mapcount(page);
 489                 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
 490                 if (mapcount >= 2)
 491                         pss /= mapcount;
 492                 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
 493                                       mapcount < 2);
 494         }
 495 }
 496
 497 #ifdef CONFIG_SHMEM
 498 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 499                           __always_unused int depth, struct mm_walk *walk)
 500 {
 501         struct mem_size_stats *mss = walk->private;
 502         struct vm_area_struct *vma = walk->vma;
 503
 504         mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping,
 505                                               linear_page_index(vma, addr),
 506                                               linear_page_index(vma, end));
 507
 508         return 0;
 509 }
 510 #else
 511 #define smaps_pte_hole          NULL
 512 #endif /* CONFIG_SHMEM */
 513
 514 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
 515 {
 516 #ifdef CONFIG_SHMEM
 517         if (walk->ops->pte_hole) {
 518                 /* depth is not used */
 519                 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
 520         }
 521 #endif
 522 }
 523
 524 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 525                 struct mm_walk *walk)
 526 {
 527         struct mem_size_stats *mss = walk->private;
 528         struct vm_area_struct *vma = walk->vma;
 529         bool locked = !!(vma->vm_flags & VM_LOCKED);
 530         struct page *page = NULL;
 531         bool migration = false, young = false, dirty = false;
 532         pte_t ptent = ptep_get(pte);
 533
 534         if (pte_present(ptent)) {
 535                 page = vm_normal_page(vma, addr, ptent);
 536                 young = pte_young(ptent);
 537                 dirty = pte_dirty(ptent);
 538         } else if (is_swap_pte(ptent)) {
 539                 swp_entry_t swpent = pte_to_swp_entry(ptent);
 540
 541                 if (!non_swap_entry(swpent)) {
 542                         int mapcount;
 543
 544                         mss->swap += PAGE_SIZE;
 545                         mapcount = swp_swapcount(swpent);
 546                         if (mapcount >= 2) {
 547                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 548
 549                                 do_div(pss_delta, mapcount);
 550                                 mss->swap_pss += pss_delta;
 551                         } else {
 552                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 553                         }
 554                 } else if (is_pfn_swap_entry(swpent)) {
 555                         if (is_migration_entry(swpent))
 556                                 migration = true;
 557                         page = pfn_swap_entry_to_page(swpent);
 558                 }
 559         } else {
 560                 smaps_pte_hole_lookup(addr, walk);
 561                 return;
 562         }
 563
 564         if (!page)
 565                 return;
 566
 567         smaps_account(mss, page, false, young, dirty, locked, migration);
 568 }
 569
 570 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 571 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 572                 struct mm_walk *walk)
 573 {
 574         struct mem_size_stats *mss = walk->private;
 575         struct vm_area_struct *vma = walk->vma;
 576         bool locked = !!(vma->vm_flags & VM_LOCKED);
 577         struct page *page = NULL;
 578         bool migration = false;
 579
 580         if (pmd_present(*pmd)) {
 581                 page = vm_normal_page_pmd(vma, addr, *pmd);
 582         } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
 583                 swp_entry_t entry = pmd_to_swp_entry(*pmd);
 584
 585                 if (is_migration_entry(entry)) {
 586                         migration = true;
 587                         page = pfn_swap_entry_to_page(entry);
 588                 }
 589         }
 590         if (IS_ERR_OR_NULL(page))
 591                 return;
 592         if (PageAnon(page))
 593                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 594         else if (PageSwapBacked(page))
 595                 mss->shmem_thp += HPAGE_PMD_SIZE;
 596         else if (is_zone_device_page(page))
 597                 /* pass */;
 598         else
 599                 mss->file_thp += HPAGE_PMD_SIZE;
 600
 601         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
 602                       locked, migration);
 603 }
 604 #else
 605 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 606                 struct mm_walk *walk)
 607 {
 608 }
 609 #endif
 610
 611 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 612                            struct mm_walk *walk)
 613 {
 614         struct vm_area_struct *vma = walk->vma;
 615         pte_t *pte;
 616         spinlock_t *ptl;
 617
 618         ptl = pmd_trans_huge_lock(pmd, vma);
 619         if (ptl) {
 620                 smaps_pmd_entry(pmd, addr, walk);
 621                 spin_unlock(ptl);
 622                 goto out;
 623         }
 624
 625         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 626         if (!pte) {
 627                 walk->action = ACTION_AGAIN;
 628                 return 0;
 629         }
 630         for (; addr != end; pte++, addr += PAGE_SIZE)
 631                 smaps_pte_entry(pte, addr, walk);
 632         pte_unmap_unlock(pte - 1, ptl);
 633 out:
 634         cond_resched();
 635         return 0;
 636 }
 637
 638 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 639 {
 640         /*
 641          * Don't forget to update Documentation/ on changes.
 642          */
 643         static const char mnemonics[BITS_PER_LONG][2] = {
 644                 /*
 645                  * In case if we meet a flag we don't know about.
 646                  */
 647                 [0 ... (BITS_PER_LONG-1)] = "??",
 648
 649                 [ilog2(VM_READ)]        = "rd",
 650                 [ilog2(VM_WRITE)]       = "wr",
 651                 [ilog2(VM_EXEC)]        = "ex",
 652                 [ilog2(VM_SHARED)]      = "sh",
 653                 [ilog2(VM_MAYREAD)]     = "mr",
 654                 [ilog2(VM_MAYWRITE)]    = "mw",
 655                 [ilog2(VM_MAYEXEC)]     = "me",
 656                 [ilog2(VM_MAYSHARE)]    = "ms",
 657                 [ilog2(VM_GROWSDOWN)]   = "gd",
 658                 [ilog2(VM_PFNMAP)]      = "pf",
 659                 [ilog2(VM_LOCKED)]      = "lo",
 660                 [ilog2(VM_IO)]          = "io",
 661                 [ilog2(VM_SEQ_READ)]    = "sr",
 662                 [ilog2(VM_RAND_READ)]   = "rr",
 663                 [ilog2(VM_DONTCOPY)]    = "dc",
 664                 [ilog2(VM_DONTEXPAND)]  = "de",
 665                 [ilog2(VM_LOCKONFAULT)] = "lf",
 666                 [ilog2(VM_ACCOUNT)]     = "ac",
 667                 [ilog2(VM_NORESERVE)]   = "nr",
 668                 [ilog2(VM_HUGETLB)]     = "ht",
 669                 [ilog2(VM_SYNC)]        = "sf",
 670                 [ilog2(VM_ARCH_1)]      = "ar",
 671                 [ilog2(VM_WIPEONFORK)]  = "wf",
 672                 [ilog2(VM_DONTDUMP)]    = "dd",
 673 #ifdef CONFIG_ARM64_BTI
 674                 [ilog2(VM_ARM64_BTI)]   = "bt",
 675 #endif
 676 #ifdef CONFIG_MEM_SOFT_DIRTY
 677                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 678 #endif
 679                 [ilog2(VM_MIXEDMAP)]    = "mm",
 680                 [ilog2(VM_HUGEPAGE)]    = "hg",
 681                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 682                 [ilog2(VM_MERGEABLE)]   = "mg",
 683                 [ilog2(VM_UFFD_MISSING)]= "um",
 684                 [ilog2(VM_UFFD_WP)]     = "uw",
 685 #ifdef CONFIG_ARM64_MTE
 686                 [ilog2(VM_MTE)]         = "mt",
 687                 [ilog2(VM_MTE_ALLOWED)] = "",
 688 #endif
 689 #ifdef CONFIG_ARCH_HAS_PKEYS
 690                 /* These come out via ProtectionKey: */
 691                 [ilog2(VM_PKEY_BIT0)]   = "",
 692                 [ilog2(VM_PKEY_BIT1)]   = "",
 693                 [ilog2(VM_PKEY_BIT2)]   = "",
 694                 [ilog2(VM_PKEY_BIT3)]   = "",
 695 #if VM_PKEY_BIT4
 696                 [ilog2(VM_PKEY_BIT4)]   = "",
 697 #endif
 698 #endif /* CONFIG_ARCH_HAS_PKEYS */
 699 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
 700                 [ilog2(VM_UFFD_MINOR)]  = "ui",
 701 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
 702 #ifdef CONFIG_X86_USER_SHADOW_STACK
 703                 [ilog2(VM_SHADOW_STACK)] = "ss",
 704 #endif
 705         };
 706         size_t i;
 707
 708         seq_puts(m, "VmFlags: ");
 709         for (i = 0; i < BITS_PER_LONG; i++) {
 710                 if (!mnemonics[i][0])
 711                         continue;
 712                 if (vma->vm_flags & (1UL << i)) {
 713                         seq_putc(m, mnemonics[i][0]);
 714                         seq_putc(m, mnemonics[i][1]);
 715                         seq_putc(m, ' ');
 716                 }
 717         }
 718         seq_putc(m, '\n');
 719 }
 720
 721 #ifdef CONFIG_HUGETLB_PAGE
 722 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 723                                  unsigned long addr, unsigned long end,
 724                                  struct mm_walk *walk)
 725 {
 726         struct mem_size_stats *mss = walk->private;
 727         struct vm_area_struct *vma = walk->vma;
 728         struct page *page = NULL;
 729         pte_t ptent = ptep_get(pte);
 730
 731         if (pte_present(ptent)) {
 732                 page = vm_normal_page(vma, addr, ptent);
 733         } else if (is_swap_pte(ptent)) {
 734                 swp_entry_t swpent = pte_to_swp_entry(ptent);
 735
 736                 if (is_pfn_swap_entry(swpent))
 737                         page = pfn_swap_entry_to_page(swpent);
 738         }
 739         if (page) {
 740                 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
 741                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 742                 else
 743                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 744         }
 745         return 0;
 746 }
 747 #else
 748 #define smaps_hugetlb_range     NULL
 749 #endif /* HUGETLB_PAGE */
 750
 751 static const struct mm_walk_ops smaps_walk_ops = {
 752         .pmd_entry              = smaps_pte_range,
 753         .hugetlb_entry          = smaps_hugetlb_range,
 754         .walk_lock              = PGWALK_RDLOCK,
 755 };
 756
 757 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 758         .pmd_entry              = smaps_pte_range,
 759         .hugetlb_entry          = smaps_hugetlb_range,
 760         .pte_hole               = smaps_pte_hole,
 761         .walk_lock              = PGWALK_RDLOCK,
 762 };
 763
 764 /*
 765  * Gather mem stats from @vma with the indicated beginning
 766  * address @start, and keep them in @mss.
 767  *
 768  * Use vm_start of @vma as the beginning address if @start is 0.
 769  */
 770 static void smap_gather_stats(struct vm_area_struct *vma,
 771                 struct mem_size_stats *mss, unsigned long start)
 772 {
 773         const struct mm_walk_ops *ops = &smaps_walk_ops;
 774
 775         /* Invalid start */
 776         if (start >= vma->vm_end)
 777                 return;
 778
 779         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 780                 /*
 781                  * For shared or readonly shmem mappings we know that all
 782                  * swapped out pages belong to the shmem object, and we can
 783                  * obtain the swap value much more efficiently. For private
 784                  * writable mappings, we might have COW pages that are
 785                  * not affected by the parent swapped out pages of the shmem
 786                  * object, so we have to distinguish them during the page walk.
 787                  * Unless we know that the shmem object (or the part mapped by
 788                  * our VMA) has no swapped out pages at all.
 789                  */
 790                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 791
 792                 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 793                                         !(vma->vm_flags & VM_WRITE))) {
 794                         mss->swap += shmem_swapped;
 795                 } else {
 796                         ops = &smaps_shmem_walk_ops;
 797                 }
 798         }
 799
 800         /* mmap_lock is held in m_start */
 801         if (!start)
 802                 walk_page_vma(vma, ops, mss);
 803         else
 804                 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
 805 }
 806
 807 #define SEQ_PUT_DEC(str, val) \
 808                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 809
 810 /* Show the contents common for smaps and smaps_rollup */
 811 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 812         bool rollup_mode)
 813 {
 814         SEQ_PUT_DEC("Rss:            ", mss->resident);
 815         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 816         SEQ_PUT_DEC(" kB\nPss_Dirty:      ", mss->pss_dirty >> PSS_SHIFT);
 817         if (rollup_mode) {
 818                 /*
 819                  * These are meaningful only for smaps_rollup, otherwise two of
 820                  * them are zero, and the other one is the same as Pss.
 821                  */
 822                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 823                         mss->pss_anon >> PSS_SHIFT);
 824                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 825                         mss->pss_file >> PSS_SHIFT);
 826                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 827                         mss->pss_shmem >> PSS_SHIFT);
 828         }
 829         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 830         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 831         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 832         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 833         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 834         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 835         SEQ_PUT_DEC(" kB\nKSM:            ", mss->ksm);
 836         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 837         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 838         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 839         SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
 840         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 841         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 842                                   mss->private_hugetlb >> 10, 7);
 843         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 844         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 845                                         mss->swap_pss >> PSS_SHIFT);
 846         SEQ_PUT_DEC(" kB\nLocked:         ",
 847                                         mss->pss_locked >> PSS_SHIFT);
 848         seq_puts(m, " kB\n");
 849 }
 850
 851 static int show_smap(struct seq_file *m, void *v)
 852 {
 853         struct vm_area_struct *vma = v;
 854         struct mem_size_stats mss;
 855
 856         memset(&mss, 0, sizeof(mss));
 857
 858         smap_gather_stats(vma, &mss, 0);
 859
 860         show_map_vma(m, vma);
 861
 862         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 863         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 864         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 865         seq_puts(m, " kB\n");
 866
 867         __show_smap(m, &mss, false);
 868
 869         seq_printf(m, "THPeligible:    %8u\n",
 870                    hugepage_vma_check(vma, vma->vm_flags, true, false, true));
 871
 872         if (arch_pkeys_enabled())
 873                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 874         show_smap_vma_flags(m, vma);
 875
 876         return 0;
 877 }
 878
 879 static int show_smaps_rollup(struct seq_file *m, void *v)
 880 {
 881         struct proc_maps_private *priv = m->private;
 882         struct mem_size_stats mss;
 883         struct mm_struct *mm = priv->mm;
 884         struct vm_area_struct *vma;
 885         unsigned long vma_start = 0, last_vma_end = 0;
 886         int ret = 0;
 887         VMA_ITERATOR(vmi, mm, 0);
 888
 889         priv->task = get_proc_task(priv->inode);
 890         if (!priv->task)
 891                 return -ESRCH;
 892
 893         if (!mm || !mmget_not_zero(mm)) {
 894                 ret = -ESRCH;
 895                 goto out_put_task;
 896         }
 897
 898         memset(&mss, 0, sizeof(mss));
 899
 900         ret = mmap_read_lock_killable(mm);
 901         if (ret)
 902                 goto out_put_mm;
 903
 904         hold_task_mempolicy(priv);
 905         vma = vma_next(&vmi);
 906
 907         if (unlikely(!vma))
 908                 goto empty_set;
 909
 910         vma_start = vma->vm_start;
 911         do {
 912                 smap_gather_stats(vma, &mss, 0);
 913                 last_vma_end = vma->vm_end;
 914
 915                 /*
 916                  * Release mmap_lock temporarily if someone wants to
 917                  * access it for write request.
 918                  */
 919                 if (mmap_lock_is_contended(mm)) {
 920                         vma_iter_invalidate(&vmi);
 921                         mmap_read_unlock(mm);
 922                         ret = mmap_read_lock_killable(mm);
 923                         if (ret) {
 924                                 release_task_mempolicy(priv);
 925                                 goto out_put_mm;
 926                         }
 927
 928                         /*
 929                          * After dropping the lock, there are four cases to
 930                          * consider. See the following example for explanation.
 931                          *
 932                          *   +------+------+-----------+
 933                          *   | VMA1 | VMA2 | VMA3      |
 934                          *   +------+------+-----------+
 935                          *   |      |      |           |
 936                          *  4k     8k     16k         400k
 937                          *
 938                          * Suppose we drop the lock after reading VMA2 due to
 939                          * contention, then we get:
 940                          *
 941                          *      last_vma_end = 16k
 942                          *
 943                          * 1) VMA2 is freed, but VMA3 exists:
 944                          *
 945                          *    vma_next(vmi) will return VMA3.
 946                          *    In this case, just continue from VMA3.
 947                          *
 948                          * 2) VMA2 still exists:
 949                          *
 950                          *    vma_next(vmi) will return VMA3.
 951                          *    In this case, just continue from VMA3.
 952                          *
 953                          * 3) No more VMAs can be found:
 954                          *
 955                          *    vma_next(vmi) will return NULL.
 956                          *    No more things to do, just break.
 957                          *
 958                          * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
 959                          *
 960                          *    vma_next(vmi) will return VMA' whose range
 961                          *    contains last_vma_end.
 962                          *    Iterate VMA' from last_vma_end.
 963                          */
 964                         vma = vma_next(&vmi);
 965                         /* Case 3 above */
 966                         if (!vma)
 967                                 break;
 968
 969                         /* Case 1 and 2 above */
 970                         if (vma->vm_start >= last_vma_end)
 971                                 continue;
 972
 973                         /* Case 4 above */
 974                         if (vma->vm_end > last_vma_end)
 975                                 smap_gather_stats(vma, &mss, last_vma_end);
 976                 }
 977         } for_each_vma(vmi, vma);
 978
 979 empty_set:
 980         show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
 981         seq_pad(m, ' ');
 982         seq_puts(m, "[rollup]\n");
 983
 984         __show_smap(m, &mss, true);
 985
 986         release_task_mempolicy(priv);
 987         mmap_read_unlock(mm);
 988
 989 out_put_mm:
 990         mmput(mm);
 991 out_put_task:
 992         put_task_struct(priv->task);
 993         priv->task = NULL;
 994
 995         return ret;
 996 }
 997 #undef SEQ_PUT_DEC
 998
 999 static const struct seq_operations proc_pid_smaps_op = {
1000         .start  = m_start,
1001         .next   = m_next,
1002         .stop   = m_stop,
1003         .show   = show_smap
1004 };
1005
1006 static int pid_smaps_open(struct inode *inode, struct file *file)
1007 {
1008         return do_maps_open(inode, file, &proc_pid_smaps_op);
1009 }
1010
1011 static int smaps_rollup_open(struct inode *inode, struct file *file)
1012 {
1013         int ret;
1014         struct proc_maps_private *priv;
1015
1016         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1017         if (!priv)
1018                 return -ENOMEM;
1019
1020         ret = single_open(file, show_smaps_rollup, priv);
1021         if (ret)
1022                 goto out_free;
1023
1024         priv->inode = inode;
1025         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1026         if (IS_ERR(priv->mm)) {
1027                 ret = PTR_ERR(priv->mm);
1028
1029                 single_release(inode, file);
1030                 goto out_free;
1031         }
1032
1033         return 0;
1034
1035 out_free:
1036         kfree(priv);
1037         return ret;
1038 }
1039
1040 static int smaps_rollup_release(struct inode *inode, struct file *file)
1041 {
1042         struct seq_file *seq = file->private_data;
1043         struct proc_maps_private *priv = seq->private;
1044
1045         if (priv->mm)
1046                 mmdrop(priv->mm);
1047
1048         kfree(priv);
1049         return single_release(inode, file);
1050 }
1051
1052 const struct file_operations proc_pid_smaps_operations = {
1053         .open           = pid_smaps_open,
1054         .read           = seq_read,
1055         .llseek         = seq_lseek,
1056         .release        = proc_map_release,
1057 };
1058
1059 const struct file_operations proc_pid_smaps_rollup_operations = {
1060         .open           = smaps_rollup_open,
1061         .read           = seq_read,
1062         .llseek         = seq_lseek,
1063         .release        = smaps_rollup_release,
1064 };
1065
1066 enum clear_refs_types {
1067         CLEAR_REFS_ALL = 1,
1068         CLEAR_REFS_ANON,
1069         CLEAR_REFS_MAPPED,
1070         CLEAR_REFS_SOFT_DIRTY,
1071         CLEAR_REFS_MM_HIWATER_RSS,
1072         CLEAR_REFS_LAST,
1073 };
1074
1075 struct clear_refs_private {
1076         enum clear_refs_types type;
1077 };
1078
1079 #ifdef CONFIG_MEM_SOFT_DIRTY
1080
1081 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1082 {
1083         struct page *page;
1084
1085         if (!pte_write(pte))
1086                 return false;
1087         if (!is_cow_mapping(vma->vm_flags))
1088                 return false;
1089         if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1090                 return false;
1091         page = vm_normal_page(vma, addr, pte);
1092         if (!page)
1093                 return false;
1094         return page_maybe_dma_pinned(page);
1095 }
1096
1097 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1098                 unsigned long addr, pte_t *pte)
1099 {
1100         /*
1101          * The soft-dirty tracker uses #PF-s to catch writes
1102          * to pages, so write-protect the pte as well. See the
1103          * Documentation/admin-guide/mm/soft-dirty.rst for full description
1104          * of how soft-dirty works.
1105          */
1106         pte_t ptent = ptep_get(pte);
1107
1108         if (pte_present(ptent)) {
1109                 pte_t old_pte;
1110
1111                 if (pte_is_pinned(vma, addr, ptent))
1112                         return;
1113                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1114                 ptent = pte_wrprotect(old_pte);
1115                 ptent = pte_clear_soft_dirty(ptent);
1116                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1117         } else if (is_swap_pte(ptent)) {
1118                 ptent = pte_swp_clear_soft_dirty(ptent);
1119                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1120         }
1121 }
1122 #else
1123 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1124                 unsigned long addr, pte_t *pte)
1125 {
1126 }
1127 #endif
1128
1129 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1130 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1131                 unsigned long addr, pmd_t *pmdp)
1132 {
1133         pmd_t old, pmd = *pmdp;
1134
1135         if (pmd_present(pmd)) {
1136                 /* See comment in change_huge_pmd() */
1137                 old = pmdp_invalidate(vma, addr, pmdp);
1138                 if (pmd_dirty(old))
1139                         pmd = pmd_mkdirty(pmd);
1140                 if (pmd_young(old))
1141                         pmd = pmd_mkyoung(pmd);
1142
1143                 pmd = pmd_wrprotect(pmd);
1144                 pmd = pmd_clear_soft_dirty(pmd);
1145
1146                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1147         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1148                 pmd = pmd_swp_clear_soft_dirty(pmd);
1149                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1150         }
1151 }
1152 #else
1153 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1154                 unsigned long addr, pmd_t *pmdp)
1155 {
1156 }
1157 #endif
1158
1159 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1160                                 unsigned long end, struct mm_walk *walk)
1161 {
1162         struct clear_refs_private *cp = walk->private;
1163         struct vm_area_struct *vma = walk->vma;
1164         pte_t *pte, ptent;
1165         spinlock_t *ptl;
1166         struct page *page;
1167
1168         ptl = pmd_trans_huge_lock(pmd, vma);
1169         if (ptl) {
1170                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1171                         clear_soft_dirty_pmd(vma, addr, pmd);
1172                         goto out;
1173                 }
1174
1175                 if (!pmd_present(*pmd))
1176                         goto out;
1177
1178                 page = pmd_page(*pmd);
1179
1180                 /* Clear accessed and referenced bits. */
1181                 pmdp_test_and_clear_young(vma, addr, pmd);
1182                 test_and_clear_page_young(page);
1183                 ClearPageReferenced(page);
1184 out:
1185                 spin_unlock(ptl);
1186                 return 0;
1187         }
1188
1189         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1190         if (!pte) {
1191                 walk->action = ACTION_AGAIN;
1192                 return 0;
1193         }
1194         for (; addr != end; pte++, addr += PAGE_SIZE) {
1195                 ptent = ptep_get(pte);
1196
1197                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1198                         clear_soft_dirty(vma, addr, pte);
1199                         continue;
1200                 }
1201
1202                 if (!pte_present(ptent))
1203                         continue;
1204
1205                 page = vm_normal_page(vma, addr, ptent);
1206                 if (!page)
1207                         continue;
1208
1209                 /* Clear accessed and referenced bits. */
1210                 ptep_test_and_clear_young(vma, addr, pte);
1211                 test_and_clear_page_young(page);
1212                 ClearPageReferenced(page);
1213         }
1214         pte_unmap_unlock(pte - 1, ptl);
1215         cond_resched();
1216         return 0;
1217 }
1218
1219 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1220                                 struct mm_walk *walk)
1221 {
1222         struct clear_refs_private *cp = walk->private;
1223         struct vm_area_struct *vma = walk->vma;
1224
1225         if (vma->vm_flags & VM_PFNMAP)
1226                 return 1;
1227
1228         /*
1229          * Writing 1 to /proc/pid/clear_refs affects all pages.
1230          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1231          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1232          * Writing 4 to /proc/pid/clear_refs affects all pages.
1233          */
1234         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1235                 return 1;
1236         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1237                 return 1;
1238         return 0;
1239 }
1240
1241 static const struct mm_walk_ops clear_refs_walk_ops = {
1242         .pmd_entry              = clear_refs_pte_range,
1243         .test_walk              = clear_refs_test_walk,
1244         .walk_lock              = PGWALK_WRLOCK,
1245 };
1246
1247 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1248                                 size_t count, loff_t *ppos)
1249 {
1250         struct task_struct *task;
1251         char buffer[PROC_NUMBUF];
1252         struct mm_struct *mm;
1253         struct vm_area_struct *vma;
1254         enum clear_refs_types type;
1255         int itype;
1256         int rv;
1257
1258         memset(buffer, 0, sizeof(buffer));
1259         if (count > sizeof(buffer) - 1)
1260                 count = sizeof(buffer) - 1;
1261         if (copy_from_user(buffer, buf, count))
1262                 return -EFAULT;
1263         rv = kstrtoint(strstrip(buffer), 10, &itype);
1264         if (rv < 0)
1265                 return rv;
1266         type = (enum clear_refs_types)itype;
1267         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1268                 return -EINVAL;
1269
1270         task = get_proc_task(file_inode(file));
1271         if (!task)
1272                 return -ESRCH;
1273         mm = get_task_mm(task);
1274         if (mm) {
1275                 VMA_ITERATOR(vmi, mm, 0);
1276                 struct mmu_notifier_range range;
1277                 struct clear_refs_private cp = {
1278                         .type = type,
1279                 };
1280
1281                 if (mmap_write_lock_killable(mm)) {
1282                         count = -EINTR;
1283                         goto out_mm;
1284                 }
1285                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1286                         /*
1287                          * Writing 5 to /proc/pid/clear_refs resets the peak
1288                          * resident set size to this mm's current rss value.
1289                          */
1290                         reset_mm_hiwater_rss(mm);
1291                         goto out_unlock;
1292                 }
1293
1294                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1295                         for_each_vma(vmi, vma) {
1296                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1297                                         continue;
1298                                 vm_flags_clear(vma, VM_SOFTDIRTY);
1299                                 vma_set_page_prot(vma);
1300                         }
1301
1302                         inc_tlb_flush_pending(mm);
1303                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1304                                                 0, mm, 0, -1UL);
1305                         mmu_notifier_invalidate_range_start(&range);
1306                 }
1307                 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1308                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1309                         mmu_notifier_invalidate_range_end(&range);
1310                         flush_tlb_mm(mm);
1311                         dec_tlb_flush_pending(mm);
1312                 }
1313 out_unlock:
1314                 mmap_write_unlock(mm);
1315 out_mm:
1316                 mmput(mm);
1317         }
1318         put_task_struct(task);
1319
1320         return count;
1321 }
1322
1323 const struct file_operations proc_clear_refs_operations = {
1324         .write          = clear_refs_write,
1325         .llseek         = noop_llseek,
1326 };
1327
1328 typedef struct {
1329         u64 pme;
1330 } pagemap_entry_t;
1331
1332 struct pagemapread {
1333         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1334         pagemap_entry_t *buffer;
1335         bool show_pfn;
1336 };
1337
1338 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1339 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1340
1341 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1342 #define PM_PFRAME_BITS          55
1343 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1344 #define PM_SOFT_DIRTY           BIT_ULL(55)
1345 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1346 #define PM_UFFD_WP              BIT_ULL(57)
1347 #define PM_FILE                 BIT_ULL(61)
1348 #define PM_SWAP                 BIT_ULL(62)
1349 #define PM_PRESENT              BIT_ULL(63)
1350
1351 #define PM_END_OF_BUFFER    1
1352
1353 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1354 {
1355         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1356 }
1357
1358 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1359                           struct pagemapread *pm)
1360 {
1361         pm->buffer[pm->pos++] = *pme;
1362         if (pm->pos >= pm->len)
1363                 return PM_END_OF_BUFFER;
1364         return 0;
1365 }
1366
1367 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1368                             __always_unused int depth, struct mm_walk *walk)
1369 {
1370         struct pagemapread *pm = walk->private;
1371         unsigned long addr = start;
1372         int err = 0;
1373
1374         while (addr < end) {
1375                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1376                 pagemap_entry_t pme = make_pme(0, 0);
1377                 /* End of address space hole, which we mark as non-present. */
1378                 unsigned long hole_end;
1379
1380                 if (vma)
1381                         hole_end = min(end, vma->vm_start);
1382                 else
1383                         hole_end = end;
1384
1385                 for (; addr < hole_end; addr += PAGE_SIZE) {
1386                         err = add_to_pagemap(addr, &pme, pm);
1387                         if (err)
1388                                 goto out;
1389                 }
1390
1391                 if (!vma)
1392                         break;
1393
1394                 /* Addresses in the VMA. */
1395                 if (vma->vm_flags & VM_SOFTDIRTY)
1396                         pme = make_pme(0, PM_SOFT_DIRTY);
1397                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1398                         err = add_to_pagemap(addr, &pme, pm);
1399                         if (err)
1400                                 goto out;
1401                 }
1402         }
1403 out:
1404         return err;
1405 }
1406
1407 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1408                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1409 {
1410         u64 frame = 0, flags = 0;
1411         struct page *page = NULL;
1412         bool migration = false;
1413
1414         if (pte_present(pte)) {
1415                 if (pm->show_pfn)
1416                         frame = pte_pfn(pte);
1417                 flags |= PM_PRESENT;
1418                 page = vm_normal_page(vma, addr, pte);
1419                 if (pte_soft_dirty(pte))
1420                         flags |= PM_SOFT_DIRTY;
1421                 if (pte_uffd_wp(pte))
1422                         flags |= PM_UFFD_WP;
1423         } else if (is_swap_pte(pte)) {
1424                 swp_entry_t entry;
1425                 if (pte_swp_soft_dirty(pte))
1426                         flags |= PM_SOFT_DIRTY;
1427                 if (pte_swp_uffd_wp(pte))
1428                         flags |= PM_UFFD_WP;
1429                 entry = pte_to_swp_entry(pte);
1430                 if (pm->show_pfn) {
1431                         pgoff_t offset;
1432                         /*
1433                          * For PFN swap offsets, keeping the offset field
1434                          * to be PFN only to be compatible with old smaps.
1435                          */
1436                         if (is_pfn_swap_entry(entry))
1437                                 offset = swp_offset_pfn(entry);
1438                         else
1439                                 offset = swp_offset(entry);
1440                         frame = swp_type(entry) |
1441                             (offset << MAX_SWAPFILES_SHIFT);
1442                 }
1443                 flags |= PM_SWAP;
1444                 migration = is_migration_entry(entry);
1445                 if (is_pfn_swap_entry(entry))
1446                         page = pfn_swap_entry_to_page(entry);
1447                 if (pte_marker_entry_uffd_wp(entry))
1448                         flags |= PM_UFFD_WP;
1449         }
1450
1451         if (page && !PageAnon(page))
1452                 flags |= PM_FILE;
1453         if (page && !migration && page_mapcount(page) == 1)
1454                 flags |= PM_MMAP_EXCLUSIVE;
1455         if (vma->vm_flags & VM_SOFTDIRTY)
1456                 flags |= PM_SOFT_DIRTY;
1457
1458         return make_pme(frame, flags);
1459 }
1460
1461 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1462                              struct mm_walk *walk)
1463 {
1464         struct vm_area_struct *vma = walk->vma;
1465         struct pagemapread *pm = walk->private;
1466         spinlock_t *ptl;
1467         pte_t *pte, *orig_pte;
1468         int err = 0;
1469 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1470         bool migration = false;
1471
1472         ptl = pmd_trans_huge_lock(pmdp, vma);
1473         if (ptl) {
1474                 u64 flags = 0, frame = 0;
1475                 pmd_t pmd = *pmdp;
1476                 struct page *page = NULL;
1477
1478                 if (vma->vm_flags & VM_SOFTDIRTY)
1479                         flags |= PM_SOFT_DIRTY;
1480
1481                 if (pmd_present(pmd)) {
1482                         page = pmd_page(pmd);
1483
1484                         flags |= PM_PRESENT;
1485                         if (pmd_soft_dirty(pmd))
1486                                 flags |= PM_SOFT_DIRTY;
1487                         if (pmd_uffd_wp(pmd))
1488                                 flags |= PM_UFFD_WP;
1489                         if (pm->show_pfn)
1490                                 frame = pmd_pfn(pmd) +
1491                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1492                 }
1493 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1494                 else if (is_swap_pmd(pmd)) {
1495                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1496                         unsigned long offset;
1497
1498                         if (pm->show_pfn) {
1499                                 if (is_pfn_swap_entry(entry))
1500                                         offset = swp_offset_pfn(entry);
1501                                 else
1502                                         offset = swp_offset(entry);
1503                                 offset = offset +
1504                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1505                                 frame = swp_type(entry) |
1506                                         (offset << MAX_SWAPFILES_SHIFT);
1507                         }
1508                         flags |= PM_SWAP;
1509                         if (pmd_swp_soft_dirty(pmd))
1510                                 flags |= PM_SOFT_DIRTY;
1511                         if (pmd_swp_uffd_wp(pmd))
1512                                 flags |= PM_UFFD_WP;
1513                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1514                         migration = is_migration_entry(entry);
1515                         page = pfn_swap_entry_to_page(entry);
1516                 }
1517 #endif
1518
1519                 if (page && !migration && page_mapcount(page) == 1)
1520                         flags |= PM_MMAP_EXCLUSIVE;
1521
1522                 for (; addr != end; addr += PAGE_SIZE) {
1523                         pagemap_entry_t pme = make_pme(frame, flags);
1524
1525                         err = add_to_pagemap(addr, &pme, pm);
1526                         if (err)
1527                                 break;
1528                         if (pm->show_pfn) {
1529                                 if (flags & PM_PRESENT)
1530                                         frame++;
1531                                 else if (flags & PM_SWAP)
1532                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1533                         }
1534                 }
1535                 spin_unlock(ptl);
1536                 return err;
1537         }
1538 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1539
1540         /*
1541          * We can assume that @vma always points to a valid one and @end never
1542          * goes beyond vma->vm_end.
1543          */
1544         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1545         if (!pte) {
1546                 walk->action = ACTION_AGAIN;
1547                 return err;
1548         }
1549         for (; addr < end; pte++, addr += PAGE_SIZE) {
1550                 pagemap_entry_t pme;
1551
1552                 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte));
1553                 err = add_to_pagemap(addr, &pme, pm);
1554                 if (err)
1555                         break;
1556         }
1557         pte_unmap_unlock(orig_pte, ptl);
1558
1559         cond_resched();
1560
1561         return err;
1562 }
1563
1564 #ifdef CONFIG_HUGETLB_PAGE
1565 /* This function walks within one hugetlb entry in the single call */
1566 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1567                                  unsigned long addr, unsigned long end,
1568                                  struct mm_walk *walk)
1569 {
1570         struct pagemapread *pm = walk->private;
1571         struct vm_area_struct *vma = walk->vma;
1572         u64 flags = 0, frame = 0;
1573         int err = 0;
1574         pte_t pte;
1575
1576         if (vma->vm_flags & VM_SOFTDIRTY)
1577                 flags |= PM_SOFT_DIRTY;
1578
1579         pte = huge_ptep_get(ptep);
1580         if (pte_present(pte)) {
1581                 struct page *page = pte_page(pte);
1582
1583                 if (!PageAnon(page))
1584                         flags |= PM_FILE;
1585
1586                 if (page_mapcount(page) == 1)
1587                         flags |= PM_MMAP_EXCLUSIVE;
1588
1589                 if (huge_pte_uffd_wp(pte))
1590                         flags |= PM_UFFD_WP;
1591
1592                 flags |= PM_PRESENT;
1593                 if (pm->show_pfn)
1594                         frame = pte_pfn(pte) +
1595                                 ((addr & ~hmask) >> PAGE_SHIFT);
1596         } else if (pte_swp_uffd_wp_any(pte)) {
1597                 flags |= PM_UFFD_WP;
1598         }
1599
1600         for (; addr != end; addr += PAGE_SIZE) {
1601                 pagemap_entry_t pme = make_pme(frame, flags);
1602
1603                 err = add_to_pagemap(addr, &pme, pm);
1604                 if (err)
1605                         return err;
1606                 if (pm->show_pfn && (flags & PM_PRESENT))
1607                         frame++;
1608         }
1609
1610         cond_resched();
1611
1612         return err;
1613 }
1614 #else
1615 #define pagemap_hugetlb_range   NULL
1616 #endif /* HUGETLB_PAGE */
1617
1618 static const struct mm_walk_ops pagemap_ops = {
1619         .pmd_entry      = pagemap_pmd_range,
1620         .pte_hole       = pagemap_pte_hole,
1621         .hugetlb_entry  = pagemap_hugetlb_range,
1622         .walk_lock      = PGWALK_RDLOCK,
1623 };
1624
1625 /*
1626  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1627  *
1628  * For each page in the address space, this file contains one 64-bit entry
1629  * consisting of the following:
1630  *
1631  * Bits 0-54  page frame number (PFN) if present
1632  * Bits 0-4   swap type if swapped
1633  * Bits 5-54  swap offset if swapped
1634  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1635  * Bit  56    page exclusively mapped
1636  * Bit  57    pte is uffd-wp write-protected
1637  * Bits 58-60 zero
1638  * Bit  61    page is file-page or shared-anon
1639  * Bit  62    page swapped
1640  * Bit  63    page present
1641  *
1642  * If the page is not present but in swap, then the PFN contains an
1643  * encoding of the swap file number and the page's offset into the
1644  * swap. Unmapped pages return a null PFN. This allows determining
1645  * precisely which pages are mapped (or in swap) and comparing mapped
1646  * pages between processes.
1647  *
1648  * Efficient users of this interface will use /proc/pid/maps to
1649  * determine which areas of memory are actually mapped and llseek to
1650  * skip over unmapped regions.
1651  */
1652 static ssize_t pagemap_read(struct file *file, char __user *buf,
1653                             size_t count, loff_t *ppos)
1654 {
1655         struct mm_struct *mm = file->private_data;
1656         struct pagemapread pm;
1657         unsigned long src;
1658         unsigned long svpfn;
1659         unsigned long start_vaddr;
1660         unsigned long end_vaddr;
1661         int ret = 0, copied = 0;
1662
1663         if (!mm || !mmget_not_zero(mm))
1664                 goto out;
1665
1666         ret = -EINVAL;
1667         /* file position must be aligned */
1668         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1669                 goto out_mm;
1670
1671         ret = 0;
1672         if (!count)
1673                 goto out_mm;
1674
1675         /* do not disclose physical addresses: attack vector */
1676         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1677
1678         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1679         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1680         ret = -ENOMEM;
1681         if (!pm.buffer)
1682                 goto out_mm;
1683
1684         src = *ppos;
1685         svpfn = src / PM_ENTRY_BYTES;
1686         end_vaddr = mm->task_size;
1687
1688         /* watch out for wraparound */
1689         start_vaddr = end_vaddr;
1690         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
1691                 unsigned long end;
1692
1693                 ret = mmap_read_lock_killable(mm);
1694                 if (ret)
1695                         goto out_free;
1696                 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
1697                 mmap_read_unlock(mm);
1698
1699                 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
1700                 if (end >= start_vaddr && end < mm->task_size)
1701                         end_vaddr = end;
1702         }
1703
1704         /* Ensure the address is inside the task */
1705         if (start_vaddr > mm->task_size)
1706                 start_vaddr = end_vaddr;
1707
1708         ret = 0;
1709         while (count && (start_vaddr < end_vaddr)) {
1710                 int len;
1711                 unsigned long end;
1712
1713                 pm.pos = 0;
1714                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1715                 /* overflow ? */
1716                 if (end < start_vaddr || end > end_vaddr)
1717                         end = end_vaddr;
1718                 ret = mmap_read_lock_killable(mm);
1719                 if (ret)
1720                         goto out_free;
1721                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1722                 mmap_read_unlock(mm);
1723                 start_vaddr = end;
1724
1725                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1726                 if (copy_to_user(buf, pm.buffer, len)) {
1727                         ret = -EFAULT;
1728                         goto out_free;
1729                 }
1730                 copied += len;
1731                 buf += len;
1732                 count -= len;
1733         }
1734         *ppos += copied;
1735         if (!ret || ret == PM_END_OF_BUFFER)
1736                 ret = copied;
1737
1738 out_free:
1739         kfree(pm.buffer);
1740 out_mm:
1741         mmput(mm);
1742 out:
1743         return ret;
1744 }
1745
1746 static int pagemap_open(struct inode *inode, struct file *file)
1747 {
1748         struct mm_struct *mm;
1749
1750         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1751         if (IS_ERR(mm))
1752                 return PTR_ERR(mm);
1753         file->private_data = mm;
1754         return 0;
1755 }
1756
1757 static int pagemap_release(struct inode *inode, struct file *file)
1758 {
1759         struct mm_struct *mm = file->private_data;
1760
1761         if (mm)
1762                 mmdrop(mm);
1763         return 0;
1764 }
1765
1766 #define PM_SCAN_CATEGORIES      (PAGE_IS_WPALLOWED | PAGE_IS_WRITTEN |  \
1767                                  PAGE_IS_FILE | PAGE_IS_PRESENT |       \
1768                                  PAGE_IS_SWAPPED | PAGE_IS_PFNZERO |    \
1769                                  PAGE_IS_HUGE)
1770 #define PM_SCAN_FLAGS           (PM_SCAN_WP_MATCHING | PM_SCAN_CHECK_WPASYNC)
1771
1772 struct pagemap_scan_private {
1773         struct pm_scan_arg arg;
1774         unsigned long masks_of_interest, cur_vma_category;
1775         struct page_region *vec_buf;
1776         unsigned long vec_buf_len, vec_buf_index, found_pages;
1777         struct page_region __user *vec_out;
1778 };
1779
1780 static unsigned long pagemap_page_category(struct pagemap_scan_private *p,
1781                                            struct vm_area_struct *vma,
1782                                            unsigned long addr, pte_t pte)
1783 {
1784         unsigned long categories = 0;
1785
1786         if (pte_present(pte)) {
1787                 struct page *page;
1788
1789                 categories |= PAGE_IS_PRESENT;
1790                 if (!pte_uffd_wp(pte))
1791                         categories |= PAGE_IS_WRITTEN;
1792
1793                 if (p->masks_of_interest & PAGE_IS_FILE) {
1794                         page = vm_normal_page(vma, addr, pte);
1795                         if (page && !PageAnon(page))
1796                                 categories |= PAGE_IS_FILE;
1797                 }
1798
1799                 if (is_zero_pfn(pte_pfn(pte)))
1800                         categories |= PAGE_IS_PFNZERO;
1801         } else if (is_swap_pte(pte)) {
1802                 swp_entry_t swp;
1803
1804                 categories |= PAGE_IS_SWAPPED;
1805                 if (!pte_swp_uffd_wp_any(pte))
1806                         categories |= PAGE_IS_WRITTEN;
1807
1808                 if (p->masks_of_interest & PAGE_IS_FILE) {
1809                         swp = pte_to_swp_entry(pte);
1810                         if (is_pfn_swap_entry(swp) &&
1811                             !PageAnon(pfn_swap_entry_to_page(swp)))
1812                                 categories |= PAGE_IS_FILE;
1813                 }
1814         }
1815
1816         return categories;
1817 }
1818
1819 static void make_uffd_wp_pte(struct vm_area_struct *vma,
1820                              unsigned long addr, pte_t *pte)
1821 {
1822         pte_t ptent = ptep_get(pte);
1823
1824         if (pte_present(ptent)) {
1825                 pte_t old_pte;
1826
1827                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1828                 ptent = pte_mkuffd_wp(ptent);
1829                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1830         } else if (is_swap_pte(ptent)) {
1831                 ptent = pte_swp_mkuffd_wp(ptent);
1832                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1833         } else {
1834                 set_pte_at(vma->vm_mm, addr, pte,
1835                            make_pte_marker(PTE_MARKER_UFFD_WP));
1836         }
1837 }
1838
1839 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1840 static unsigned long pagemap_thp_category(struct pagemap_scan_private *p,
1841                                           struct vm_area_struct *vma,
1842                                           unsigned long addr, pmd_t pmd)
1843 {
1844         unsigned long categories = PAGE_IS_HUGE;
1845
1846         if (pmd_present(pmd)) {
1847                 struct page *page;
1848
1849                 categories |= PAGE_IS_PRESENT;
1850                 if (!pmd_uffd_wp(pmd))
1851                         categories |= PAGE_IS_WRITTEN;
1852
1853                 if (p->masks_of_interest & PAGE_IS_FILE) {
1854                         page = vm_normal_page_pmd(vma, addr, pmd);
1855                         if (page && !PageAnon(page))
1856                                 categories |= PAGE_IS_FILE;
1857                 }
1858
1859                 if (is_zero_pfn(pmd_pfn(pmd)))
1860                         categories |= PAGE_IS_PFNZERO;
1861         } else if (is_swap_pmd(pmd)) {
1862                 swp_entry_t swp;
1863
1864                 categories |= PAGE_IS_SWAPPED;
1865                 if (!pmd_swp_uffd_wp(pmd))
1866                         categories |= PAGE_IS_WRITTEN;
1867
1868                 if (p->masks_of_interest & PAGE_IS_FILE) {
1869                         swp = pmd_to_swp_entry(pmd);
1870                         if (is_pfn_swap_entry(swp) &&
1871                             !PageAnon(pfn_swap_entry_to_page(swp)))
1872                                 categories |= PAGE_IS_FILE;
1873                 }
1874         }
1875
1876         return categories;
1877 }
1878
1879 static void make_uffd_wp_pmd(struct vm_area_struct *vma,
1880                              unsigned long addr, pmd_t *pmdp)
1881 {
1882         pmd_t old, pmd = *pmdp;
1883
1884         if (pmd_present(pmd)) {
1885                 old = pmdp_invalidate_ad(vma, addr, pmdp);
1886                 pmd = pmd_mkuffd_wp(old);
1887                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1888         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1889                 pmd = pmd_swp_mkuffd_wp(pmd);
1890                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1891         }
1892 }
1893 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1894
1895 #ifdef CONFIG_HUGETLB_PAGE
1896 static unsigned long pagemap_hugetlb_category(pte_t pte)
1897 {
1898         unsigned long categories = PAGE_IS_HUGE;
1899
1900         /*
1901          * According to pagemap_hugetlb_range(), file-backed HugeTLB
1902          * page cannot be swapped. So PAGE_IS_FILE is not checked for
1903          * swapped pages.
1904          */
1905         if (pte_present(pte)) {
1906                 categories |= PAGE_IS_PRESENT;
1907                 if (!huge_pte_uffd_wp(pte))
1908                         categories |= PAGE_IS_WRITTEN;
1909                 if (!PageAnon(pte_page(pte)))
1910                         categories |= PAGE_IS_FILE;
1911                 if (is_zero_pfn(pte_pfn(pte)))
1912                         categories |= PAGE_IS_PFNZERO;
1913         } else if (is_swap_pte(pte)) {
1914                 categories |= PAGE_IS_SWAPPED;
1915                 if (!pte_swp_uffd_wp_any(pte))
1916                         categories |= PAGE_IS_WRITTEN;
1917         }
1918
1919         return categories;
1920 }
1921
1922 static void make_uffd_wp_huge_pte(struct vm_area_struct *vma,
1923                                   unsigned long addr, pte_t *ptep,
1924                                   pte_t ptent)
1925 {
1926         unsigned long psize;
1927
1928         if (is_hugetlb_entry_hwpoisoned(ptent) || is_pte_marker(ptent))
1929                 return;
1930
1931         psize = huge_page_size(hstate_vma(vma));
1932
1933         if (is_hugetlb_entry_migration(ptent))
1934                 set_huge_pte_at(vma->vm_mm, addr, ptep,
1935                                 pte_swp_mkuffd_wp(ptent), psize);
1936         else if (!huge_pte_none(ptent))
1937                 huge_ptep_modify_prot_commit(vma, addr, ptep, ptent,
1938                                              huge_pte_mkuffd_wp(ptent));
1939         else
1940                 set_huge_pte_at(vma->vm_mm, addr, ptep,
1941                                 make_pte_marker(PTE_MARKER_UFFD_WP), psize);
1942 }
1943 #endif /* CONFIG_HUGETLB_PAGE */
1944
1945 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE)
1946 static void pagemap_scan_backout_range(struct pagemap_scan_private *p,
1947                                        unsigned long addr, unsigned long end)
1948 {
1949         struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
1950
1951         if (cur_buf->start != addr)
1952                 cur_buf->end = addr;
1953         else
1954                 cur_buf->start = cur_buf->end = 0;
1955
1956         p->found_pages -= (end - addr) / PAGE_SIZE;
1957 }
1958 #endif
1959
1960 static bool pagemap_scan_is_interesting_page(unsigned long categories,
1961                                              const struct pagemap_scan_private *p)
1962 {
1963         categories ^= p->arg.category_inverted;
1964         if ((categories & p->arg.category_mask) != p->arg.category_mask)
1965                 return false;
1966         if (p->arg.category_anyof_mask && !(categories & p->arg.category_anyof_mask))
1967                 return false;
1968
1969         return true;
1970 }
1971
1972 static bool pagemap_scan_is_interesting_vma(unsigned long categories,
1973                                             const struct pagemap_scan_private *p)
1974 {
1975         unsigned long required = p->arg.category_mask & PAGE_IS_WPALLOWED;
1976
1977         categories ^= p->arg.category_inverted;
1978         if ((categories & required) != required)
1979                 return false;
1980
1981         return true;
1982 }
1983
1984 static int pagemap_scan_test_walk(unsigned long start, unsigned long end,
1985                                   struct mm_walk *walk)
1986 {
1987         struct pagemap_scan_private *p = walk->private;
1988         struct vm_area_struct *vma = walk->vma;
1989         unsigned long vma_category = 0;
1990
1991         if (userfaultfd_wp_async(vma) && userfaultfd_wp_use_markers(vma))
1992                 vma_category |= PAGE_IS_WPALLOWED;
1993         else if (p->arg.flags & PM_SCAN_CHECK_WPASYNC)
1994                 return -EPERM;
1995
1996         if (vma->vm_flags & VM_PFNMAP)
1997                 return 1;
1998
1999         if (!pagemap_scan_is_interesting_vma(vma_category, p))
2000                 return 1;
2001
2002         p->cur_vma_category = vma_category;
2003
2004         return 0;
2005 }
2006
2007 static bool pagemap_scan_push_range(unsigned long categories,
2008                                     struct pagemap_scan_private *p,
2009                                     unsigned long addr, unsigned long end)
2010 {
2011         struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
2012
2013         /*
2014          * When there is no output buffer provided at all, the sentinel values
2015          * won't match here. There is no other way for `cur_buf->end` to be
2016          * non-zero other than it being non-empty.
2017          */
2018         if (addr == cur_buf->end && categories == cur_buf->categories) {
2019                 cur_buf->end = end;
2020                 return true;
2021         }
2022
2023         if (cur_buf->end) {
2024                 if (p->vec_buf_index >= p->vec_buf_len - 1)
2025                         return false;
2026
2027                 cur_buf = &p->vec_buf[++p->vec_buf_index];
2028         }
2029
2030         cur_buf->start = addr;
2031         cur_buf->end = end;
2032         cur_buf->categories = categories;
2033
2034         return true;
2035 }
2036
2037 static int pagemap_scan_output(unsigned long categories,
2038                                struct pagemap_scan_private *p,
2039                                unsigned long addr, unsigned long *end)
2040 {
2041         unsigned long n_pages, total_pages;
2042         int ret = 0;
2043
2044         if (!p->vec_buf)
2045                 return 0;
2046
2047         categories &= p->arg.return_mask;
2048
2049         n_pages = (*end - addr) / PAGE_SIZE;
2050         if (check_add_overflow(p->found_pages, n_pages, &total_pages) ||
2051             total_pages > p->arg.max_pages) {
2052                 size_t n_too_much = total_pages - p->arg.max_pages;
2053                 *end -= n_too_much * PAGE_SIZE;
2054                 n_pages -= n_too_much;
2055                 ret = -ENOSPC;
2056         }
2057
2058         if (!pagemap_scan_push_range(categories, p, addr, *end)) {
2059                 *end = addr;
2060                 n_pages = 0;
2061                 ret = -ENOSPC;
2062         }
2063
2064         p->found_pages += n_pages;
2065         if (ret)
2066                 p->arg.walk_end = *end;
2067
2068         return ret;
2069 }
2070
2071 static int pagemap_scan_thp_entry(pmd_t *pmd, unsigned long start,
2072                                   unsigned long end, struct mm_walk *walk)
2073 {
2074 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2075         struct pagemap_scan_private *p = walk->private;
2076         struct vm_area_struct *vma = walk->vma;
2077         unsigned long categories;
2078         spinlock_t *ptl;
2079         int ret = 0;
2080
2081         ptl = pmd_trans_huge_lock(pmd, vma);
2082         if (!ptl)
2083                 return -ENOENT;
2084
2085         categories = p->cur_vma_category |
2086                      pagemap_thp_category(p, vma, start, *pmd);
2087
2088         if (!pagemap_scan_is_interesting_page(categories, p))
2089                 goto out_unlock;
2090
2091         ret = pagemap_scan_output(categories, p, start, &end);
2092         if (start == end)
2093                 goto out_unlock;
2094
2095         if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2096                 goto out_unlock;
2097         if (~categories & PAGE_IS_WRITTEN)
2098                 goto out_unlock;
2099
2100         /*
2101          * Break huge page into small pages if the WP operation
2102          * needs to be performed on a portion of the huge page.
2103          */
2104         if (end != start + HPAGE_SIZE) {
2105                 spin_unlock(ptl);
2106                 split_huge_pmd(vma, pmd, start);
2107                 pagemap_scan_backout_range(p, start, end);
2108                 /* Report as if there was no THP */
2109                 return -ENOENT;
2110         }
2111
2112         make_uffd_wp_pmd(vma, start, pmd);
2113         flush_tlb_range(vma, start, end);
2114 out_unlock:
2115         spin_unlock(ptl);
2116         return ret;
2117 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
2118         return -ENOENT;
2119 #endif
2120 }
2121
2122 static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start,
2123                                   unsigned long end, struct mm_walk *walk)
2124 {
2125         struct pagemap_scan_private *p = walk->private;
2126         struct vm_area_struct *vma = walk->vma;
2127         unsigned long addr, flush_end = 0;
2128         pte_t *pte, *start_pte;
2129         spinlock_t *ptl;
2130         int ret;
2131
2132         arch_enter_lazy_mmu_mode();
2133
2134         ret = pagemap_scan_thp_entry(pmd, start, end, walk);
2135         if (ret != -ENOENT) {
2136                 arch_leave_lazy_mmu_mode();
2137                 return ret;
2138         }
2139
2140         ret = 0;
2141         start_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl);
2142         if (!pte) {
2143                 arch_leave_lazy_mmu_mode();
2144                 walk->action = ACTION_AGAIN;
2145                 return 0;
2146         }
2147
2148         if (!p->vec_out) {
2149                 /* Fast path for performing exclusive WP */
2150                 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2151                         if (pte_uffd_wp(ptep_get(pte)))
2152                                 continue;
2153                         make_uffd_wp_pte(vma, addr, pte);
2154                         if (!flush_end)
2155                                 start = addr;
2156                         flush_end = addr + PAGE_SIZE;
2157                 }
2158                 goto flush_and_return;
2159         }
2160
2161         if (!p->arg.category_anyof_mask && !p->arg.category_inverted &&
2162             p->arg.category_mask == PAGE_IS_WRITTEN &&
2163             p->arg.return_mask == PAGE_IS_WRITTEN) {
2164                 for (addr = start; addr < end; pte++, addr += PAGE_SIZE) {
2165                         unsigned long next = addr + PAGE_SIZE;
2166
2167                         if (pte_uffd_wp(ptep_get(pte)))
2168                                 continue;
2169                         ret = pagemap_scan_output(p->cur_vma_category | PAGE_IS_WRITTEN,
2170                                                   p, addr, &next);
2171                         if (next == addr)
2172                                 break;
2173                         if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2174                                 continue;
2175                         make_uffd_wp_pte(vma, addr, pte);
2176                         if (!flush_end)
2177                                 start = addr;
2178                         flush_end = next;
2179                 }
2180                 goto flush_and_return;
2181         }
2182
2183         for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2184                 unsigned long categories = p->cur_vma_category |
2185                                            pagemap_page_category(p, vma, addr, ptep_get(pte));
2186                 unsigned long next = addr + PAGE_SIZE;
2187
2188                 if (!pagemap_scan_is_interesting_page(categories, p))
2189                         continue;
2190
2191                 ret = pagemap_scan_output(categories, p, addr, &next);
2192                 if (next == addr)
2193                         break;
2194
2195                 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2196                         continue;
2197                 if (~categories & PAGE_IS_WRITTEN)
2198                         continue;
2199
2200                 make_uffd_wp_pte(vma, addr, pte);
2201                 if (!flush_end)
2202                         start = addr;
2203                 flush_end = next;
2204         }
2205
2206 flush_and_return:
2207         if (flush_end)
2208                 flush_tlb_range(vma, start, addr);
2209
2210         pte_unmap_unlock(start_pte, ptl);
2211         arch_leave_lazy_mmu_mode();
2212
2213         cond_resched();
2214         return ret;
2215 }
2216
2217 #ifdef CONFIG_HUGETLB_PAGE
2218 static int pagemap_scan_hugetlb_entry(pte_t *ptep, unsigned long hmask,
2219                                       unsigned long start, unsigned long end,
2220                                       struct mm_walk *walk)
2221 {
2222         struct pagemap_scan_private *p = walk->private;
2223         struct vm_area_struct *vma = walk->vma;
2224         unsigned long categories;
2225         spinlock_t *ptl;
2226         int ret = 0;
2227         pte_t pte;
2228
2229         if (~p->arg.flags & PM_SCAN_WP_MATCHING) {
2230                 /* Go the short route when not write-protecting pages. */
2231
2232                 pte = huge_ptep_get(ptep);
2233                 categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2234
2235                 if (!pagemap_scan_is_interesting_page(categories, p))
2236                         return 0;
2237
2238                 return pagemap_scan_output(categories, p, start, &end);
2239         }
2240
2241         i_mmap_lock_write(vma->vm_file->f_mapping);
2242         ptl = huge_pte_lock(hstate_vma(vma), vma->vm_mm, ptep);
2243
2244         pte = huge_ptep_get(ptep);
2245         categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2246
2247         if (!pagemap_scan_is_interesting_page(categories, p))
2248                 goto out_unlock;
2249
2250         ret = pagemap_scan_output(categories, p, start, &end);
2251         if (start == end)
2252                 goto out_unlock;
2253
2254         if (~categories & PAGE_IS_WRITTEN)
2255                 goto out_unlock;
2256
2257         if (end != start + HPAGE_SIZE) {
2258                 /* Partial HugeTLB page WP isn't possible. */
2259                 pagemap_scan_backout_range(p, start, end);
2260                 p->arg.walk_end = start;
2261                 ret = 0;
2262                 goto out_unlock;
2263         }
2264
2265         make_uffd_wp_huge_pte(vma, start, ptep, pte);
2266         flush_hugetlb_tlb_range(vma, start, end);
2267
2268 out_unlock:
2269         spin_unlock(ptl);
2270         i_mmap_unlock_write(vma->vm_file->f_mapping);
2271
2272         return ret;
2273 }
2274 #else
2275 #define pagemap_scan_hugetlb_entry NULL
2276 #endif
2277
2278 static int pagemap_scan_pte_hole(unsigned long addr, unsigned long end,
2279                                  int depth, struct mm_walk *walk)
2280 {
2281         struct pagemap_scan_private *p = walk->private;
2282         struct vm_area_struct *vma = walk->vma;
2283         int ret, err;
2284
2285         if (!vma || !pagemap_scan_is_interesting_page(p->cur_vma_category, p))
2286                 return 0;
2287
2288         ret = pagemap_scan_output(p->cur_vma_category, p, addr, &end);
2289         if (addr == end)
2290                 return ret;
2291
2292         if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2293                 return ret;
2294
2295         err = uffd_wp_range(vma, addr, end - addr, true);
2296         if (err < 0)
2297                 ret = err;
2298
2299         return ret;
2300 }
2301
2302 static const struct mm_walk_ops pagemap_scan_ops = {
2303         .test_walk = pagemap_scan_test_walk,
2304         .pmd_entry = pagemap_scan_pmd_entry,
2305         .pte_hole = pagemap_scan_pte_hole,
2306         .hugetlb_entry = pagemap_scan_hugetlb_entry,
2307 };
2308
2309 static int pagemap_scan_get_args(struct pm_scan_arg *arg,
2310                                  unsigned long uarg)
2311 {
2312         if (copy_from_user(arg, (void __user *)uarg, sizeof(*arg)))
2313                 return -EFAULT;
2314
2315         if (arg->size != sizeof(struct pm_scan_arg))
2316                 return -EINVAL;
2317
2318         /* Validate requested features */
2319         if (arg->flags & ~PM_SCAN_FLAGS)
2320                 return -EINVAL;
2321         if ((arg->category_inverted | arg->category_mask |
2322              arg->category_anyof_mask | arg->return_mask) & ~PM_SCAN_CATEGORIES)
2323                 return -EINVAL;
2324
2325         arg->start = untagged_addr((unsigned long)arg->start);
2326         arg->end = untagged_addr((unsigned long)arg->end);
2327         arg->vec = untagged_addr((unsigned long)arg->vec);
2328
2329         /* Validate memory pointers */
2330         if (!IS_ALIGNED(arg->start, PAGE_SIZE))
2331                 return -EINVAL;
2332         if (!access_ok((void __user *)(long)arg->start, arg->end - arg->start))
2333                 return -EFAULT;
2334         if (!arg->vec && arg->vec_len)
2335                 return -EINVAL;
2336         if (arg->vec && !access_ok((void __user *)(long)arg->vec,
2337                               arg->vec_len * sizeof(struct page_region)))
2338                 return -EFAULT;
2339
2340         /* Fixup default values */
2341         arg->end = ALIGN(arg->end, PAGE_SIZE);
2342         arg->walk_end = 0;
2343         if (!arg->max_pages)
2344                 arg->max_pages = ULONG_MAX;
2345
2346         return 0;
2347 }
2348
2349 static int pagemap_scan_writeback_args(struct pm_scan_arg *arg,
2350                                        unsigned long uargl)
2351 {
2352         struct pm_scan_arg __user *uarg = (void __user *)uargl;
2353
2354         if (copy_to_user(&uarg->walk_end, &arg->walk_end, sizeof(arg->walk_end)))
2355                 return -EFAULT;
2356
2357         return 0;
2358 }
2359
2360 static int pagemap_scan_init_bounce_buffer(struct pagemap_scan_private *p)
2361 {
2362         if (!p->arg.vec_len)
2363                 return 0;
2364
2365         p->vec_buf_len = min_t(size_t, PAGEMAP_WALK_SIZE >> PAGE_SHIFT,
2366                                p->arg.vec_len);
2367         p->vec_buf = kmalloc_array(p->vec_buf_len, sizeof(*p->vec_buf),
2368                                    GFP_KERNEL);
2369         if (!p->vec_buf)
2370                 return -ENOMEM;
2371
2372         p->vec_buf->start = p->vec_buf->end = 0;
2373         p->vec_out = (struct page_region __user *)(long)p->arg.vec;
2374
2375         return 0;
2376 }
2377
2378 static long pagemap_scan_flush_buffer(struct pagemap_scan_private *p)
2379 {
2380         const struct page_region *buf = p->vec_buf;
2381         long n = p->vec_buf_index;
2382
2383         if (!p->vec_buf)
2384                 return 0;
2385
2386         if (buf[n].end != buf[n].start)
2387                 n++;
2388
2389         if (!n)
2390                 return 0;
2391
2392         if (copy_to_user(p->vec_out, buf, n * sizeof(*buf)))
2393                 return -EFAULT;
2394
2395         p->arg.vec_len -= n;
2396         p->vec_out += n;
2397
2398         p->vec_buf_index = 0;
2399         p->vec_buf_len = min_t(size_t, p->vec_buf_len, p->arg.vec_len);
2400         p->vec_buf->start = p->vec_buf->end = 0;
2401
2402         return n;
2403 }
2404
2405 static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg)
2406 {
2407         struct mmu_notifier_range range;
2408         struct pagemap_scan_private p = {0};
2409         unsigned long walk_start;
2410         size_t n_ranges_out = 0;
2411         int ret;
2412
2413         ret = pagemap_scan_get_args(&p.arg, uarg);
2414         if (ret)
2415                 return ret;
2416
2417         p.masks_of_interest = p.arg.category_mask | p.arg.category_anyof_mask |
2418                               p.arg.return_mask;
2419         ret = pagemap_scan_init_bounce_buffer(&p);
2420         if (ret)
2421                 return ret;
2422
2423         /* Protection change for the range is going to happen. */
2424         if (p.arg.flags & PM_SCAN_WP_MATCHING) {
2425                 mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_VMA, 0,
2426                                         mm, p.arg.start, p.arg.end);
2427                 mmu_notifier_invalidate_range_start(&range);
2428         }
2429
2430         for (walk_start = p.arg.start; walk_start < p.arg.end;
2431                         walk_start = p.arg.walk_end) {
2432                 long n_out;
2433
2434                 if (fatal_signal_pending(current)) {
2435                         ret = -EINTR;
2436                         break;
2437                 }
2438
2439                 ret = mmap_read_lock_killable(mm);
2440                 if (ret)
2441                         break;
2442                 ret = walk_page_range(mm, walk_start, p.arg.end,
2443                                       &pagemap_scan_ops, &p);
2444                 mmap_read_unlock(mm);
2445
2446                 n_out = pagemap_scan_flush_buffer(&p);
2447                 if (n_out < 0)
2448                         ret = n_out;
2449                 else
2450                         n_ranges_out += n_out;
2451
2452                 if (ret != -ENOSPC)
2453                         break;
2454
2455                 if (p.arg.vec_len == 0 || p.found_pages == p.arg.max_pages)
2456                         break;
2457         }
2458
2459         /* ENOSPC signifies early stop (buffer full) from the walk. */
2460         if (!ret || ret == -ENOSPC)
2461                 ret = n_ranges_out;
2462
2463         /* The walk_end isn't set when ret is zero */
2464         if (!p.arg.walk_end)
2465                 p.arg.walk_end = p.arg.end;
2466         if (pagemap_scan_writeback_args(&p.arg, uarg))
2467                 ret = -EFAULT;
2468
2469         if (p.arg.flags & PM_SCAN_WP_MATCHING)
2470                 mmu_notifier_invalidate_range_end(&range);
2471
2472         kfree(p.vec_buf);
2473         return ret;
2474 }
2475
2476 static long do_pagemap_cmd(struct file *file, unsigned int cmd,
2477                            unsigned long arg)
2478 {
2479         struct mm_struct *mm = file->private_data;
2480
2481         switch (cmd) {
2482         case PAGEMAP_SCAN:
2483                 return do_pagemap_scan(mm, arg);
2484
2485         default:
2486                 return -EINVAL;
2487         }
2488 }
2489
2490 const struct file_operations proc_pagemap_operations = {
2491         .llseek         = mem_lseek, /* borrow this */
2492         .read           = pagemap_read,
2493         .open           = pagemap_open,
2494         .release        = pagemap_release,
2495         .unlocked_ioctl = do_pagemap_cmd,
2496         .compat_ioctl   = do_pagemap_cmd,
2497 };
2498 #endif /* CONFIG_PROC_PAGE_MONITOR */
2499
2500 #ifdef CONFIG_NUMA
2501
2502 struct numa_maps {
2503         unsigned long pages;
2504         unsigned long anon;
2505         unsigned long active;
2506         unsigned long writeback;
2507         unsigned long mapcount_max;
2508         unsigned long dirty;
2509         unsigned long swapcache;
2510         unsigned long node[MAX_NUMNODES];
2511 };
2512
2513 struct numa_maps_private {
2514         struct proc_maps_private proc_maps;
2515         struct numa_maps md;
2516 };
2517
2518 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
2519                         unsigned long nr_pages)
2520 {
2521         int count = page_mapcount(page);
2522
2523         md->pages += nr_pages;
2524         if (pte_dirty || PageDirty(page))
2525                 md->dirty += nr_pages;
2526
2527         if (PageSwapCache(page))
2528                 md->swapcache += nr_pages;
2529
2530         if (PageActive(page) || PageUnevictable(page))
2531                 md->active += nr_pages;
2532
2533         if (PageWriteback(page))
2534                 md->writeback += nr_pages;
2535
2536         if (PageAnon(page))
2537                 md->anon += nr_pages;
2538
2539         if (count > md->mapcount_max)
2540                 md->mapcount_max = count;
2541
2542         md->node[page_to_nid(page)] += nr_pages;
2543 }
2544
2545 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
2546                 unsigned long addr)
2547 {
2548         struct page *page;
2549         int nid;
2550
2551         if (!pte_present(pte))
2552                 return NULL;
2553
2554         page = vm_normal_page(vma, addr, pte);
2555         if (!page || is_zone_device_page(page))
2556                 return NULL;
2557
2558         if (PageReserved(page))
2559                 return NULL;
2560
2561         nid = page_to_nid(page);
2562         if (!node_isset(nid, node_states[N_MEMORY]))
2563                 return NULL;
2564
2565         return page;
2566 }
2567
2568 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2569 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
2570                                               struct vm_area_struct *vma,
2571                                               unsigned long addr)
2572 {
2573         struct page *page;
2574         int nid;
2575
2576         if (!pmd_present(pmd))
2577                 return NULL;
2578
2579         page = vm_normal_page_pmd(vma, addr, pmd);
2580         if (!page)
2581                 return NULL;
2582
2583         if (PageReserved(page))
2584                 return NULL;
2585
2586         nid = page_to_nid(page);
2587         if (!node_isset(nid, node_states[N_MEMORY]))
2588                 return NULL;
2589
2590         return page;
2591 }
2592 #endif
2593
2594 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
2595                 unsigned long end, struct mm_walk *walk)
2596 {
2597         struct numa_maps *md = walk->private;
2598         struct vm_area_struct *vma = walk->vma;
2599         spinlock_t *ptl;
2600         pte_t *orig_pte;
2601         pte_t *pte;
2602
2603 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2604         ptl = pmd_trans_huge_lock(pmd, vma);
2605         if (ptl) {
2606                 struct page *page;
2607
2608                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
2609                 if (page)
2610                         gather_stats(page, md, pmd_dirty(*pmd),
2611                                      HPAGE_PMD_SIZE/PAGE_SIZE);
2612                 spin_unlock(ptl);
2613                 return 0;
2614         }
2615 #endif
2616         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
2617         if (!pte) {
2618                 walk->action = ACTION_AGAIN;
2619                 return 0;
2620         }
2621         do {
2622                 pte_t ptent = ptep_get(pte);
2623                 struct page *page = can_gather_numa_stats(ptent, vma, addr);
2624                 if (!page)
2625                         continue;
2626                 gather_stats(page, md, pte_dirty(ptent), 1);
2627
2628         } while (pte++, addr += PAGE_SIZE, addr != end);
2629         pte_unmap_unlock(orig_pte, ptl);
2630         cond_resched();
2631         return 0;
2632 }
2633 #ifdef CONFIG_HUGETLB_PAGE
2634 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
2635                 unsigned long addr, unsigned long end, struct mm_walk *walk)
2636 {
2637         pte_t huge_pte = huge_ptep_get(pte);
2638         struct numa_maps *md;
2639         struct page *page;
2640
2641         if (!pte_present(huge_pte))
2642                 return 0;
2643
2644         page = pte_page(huge_pte);
2645
2646         md = walk->private;
2647         gather_stats(page, md, pte_dirty(huge_pte), 1);
2648         return 0;
2649 }
2650
2651 #else
2652 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
2653                 unsigned long addr, unsigned long end, struct mm_walk *walk)
2654 {
2655         return 0;
2656 }
2657 #endif
2658
2659 static const struct mm_walk_ops show_numa_ops = {
2660         .hugetlb_entry = gather_hugetlb_stats,
2661         .pmd_entry = gather_pte_stats,
2662         .walk_lock = PGWALK_RDLOCK,
2663 };
2664
2665 /*
2666  * Display pages allocated per node and memory policy via /proc.
2667  */
2668 static int show_numa_map(struct seq_file *m, void *v)
2669 {
2670         struct numa_maps_private *numa_priv = m->private;
2671         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
2672         struct vm_area_struct *vma = v;
2673         struct numa_maps *md = &numa_priv->md;
2674         struct file *file = vma->vm_file;
2675         struct mm_struct *mm = vma->vm_mm;
2676         char buffer[64];
2677         struct mempolicy *pol;
2678         pgoff_t ilx;
2679         int nid;
2680
2681         if (!mm)
2682                 return 0;
2683
2684         /* Ensure we start with an empty set of numa_maps statistics. */
2685         memset(md, 0, sizeof(*md));
2686
2687         pol = __get_vma_policy(vma, vma->vm_start, &ilx);
2688         if (pol) {
2689                 mpol_to_str(buffer, sizeof(buffer), pol);
2690                 mpol_cond_put(pol);
2691         } else {
2692                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
2693         }
2694
2695         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
2696
2697         if (file) {
2698                 seq_puts(m, " file=");
2699                 seq_path(m, file_user_path(file), "\n\t= ");
2700         } else if (vma_is_initial_heap(vma)) {
2701                 seq_puts(m, " heap");
2702         } else if (vma_is_initial_stack(vma)) {
2703                 seq_puts(m, " stack");
2704         }
2705
2706         if (is_vm_hugetlb_page(vma))
2707                 seq_puts(m, " huge");
2708
2709         /* mmap_lock is held by m_start */
2710         walk_page_vma(vma, &show_numa_ops, md);
2711
2712         if (!md->pages)
2713                 goto out;
2714
2715         if (md->anon)
2716                 seq_printf(m, " anon=%lu", md->anon);
2717
2718         if (md->dirty)
2719                 seq_printf(m, " dirty=%lu", md->dirty);
2720
2721         if (md->pages != md->anon && md->pages != md->dirty)
2722                 seq_printf(m, " mapped=%lu", md->pages);
2723
2724         if (md->mapcount_max > 1)
2725                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
2726
2727         if (md->swapcache)
2728                 seq_printf(m, " swapcache=%lu", md->swapcache);
2729
2730         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2731                 seq_printf(m, " active=%lu", md->active);
2732
2733         if (md->writeback)
2734                 seq_printf(m, " writeback=%lu", md->writeback);
2735
2736         for_each_node_state(nid, N_MEMORY)
2737                 if (md->node[nid])
2738                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2739
2740         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2741 out:
2742         seq_putc(m, '\n');
2743         return 0;
2744 }
2745
2746 static const struct seq_operations proc_pid_numa_maps_op = {
2747         .start  = m_start,
2748         .next   = m_next,
2749         .stop   = m_stop,
2750         .show   = show_numa_map,
2751 };
2752
2753 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2754 {
2755         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2756                                 sizeof(struct numa_maps_private));
2757 }
2758
2759 const struct file_operations proc_pid_numa_maps_operations = {
2760         .open           = pid_numa_maps_open,
2761         .read           = seq_read,
2762         .llseek         = seq_lseek,
2763         .release        = proc_map_release,
2764 };
2765
2766 #endif /* CONFIG_NUMA */