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                 const struct inode *inode = file_user_inode(vma->vm_file);
 277
 278                 dev = inode->i_sb->s_dev;
 279                 ino = inode->i_ino;
 280                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 281         }
 282
 283         start = vma->vm_start;
 284         end = vma->vm_end;
 285         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 286         if (mm)
 287                 anon_name = anon_vma_name(vma);
 288
 289         /*
 290          * Print the dentry name for named mappings, and a
 291          * special [heap] marker for the heap:
 292          */
 293         if (file) {
 294                 seq_pad(m, ' ');
 295                 /*
 296                  * If user named this anon shared memory via
 297                  * prctl(PR_SET_VMA ..., use the provided name.
 298                  */
 299                 if (anon_name)
 300                         seq_printf(m, "[anon_shmem:%s]", anon_name->name);
 301                 else
 302                         seq_path(m, file_user_path(file), "\n");
 303                 goto done;
 304         }
 305
 306         if (vma->vm_ops && vma->vm_ops->name) {
 307                 name = vma->vm_ops->name(vma);
 308                 if (name)
 309                         goto done;
 310         }
 311
 312         name = arch_vma_name(vma);
 313         if (!name) {
 314                 if (!mm) {
 315                         name = "[vdso]";
 316                         goto done;
 317                 }
 318
 319                 if (vma_is_initial_heap(vma)) {
 320                         name = "[heap]";
 321                         goto done;
 322                 }
 323
 324                 if (vma_is_initial_stack(vma)) {
 325                         name = "[stack]";
 326                         goto done;
 327                 }
 328
 329                 if (anon_name) {
 330                         seq_pad(m, ' ');
 331                         seq_printf(m, "[anon:%s]", anon_name->name);
 332                 }
 333         }
 334
 335 done:
 336         if (name) {
 337                 seq_pad(m, ' ');
 338                 seq_puts(m, name);
 339         }
 340         seq_putc(m, '\n');
 341 }
 342
 343 static int show_map(struct seq_file *m, void *v)
 344 {
 345         show_map_vma(m, v);
 346         return 0;
 347 }
 348
 349 static const struct seq_operations proc_pid_maps_op = {
 350         .start  = m_start,
 351         .next   = m_next,
 352         .stop   = m_stop,
 353         .show   = show_map
 354 };
 355
 356 static int pid_maps_open(struct inode *inode, struct file *file)
 357 {
 358         return do_maps_open(inode, file, &proc_pid_maps_op);
 359 }
 360
 361 const struct file_operations proc_pid_maps_operations = {
 362         .open           = pid_maps_open,
 363         .read           = seq_read,
 364         .llseek         = seq_lseek,
 365         .release        = proc_map_release,
 366 };
 367
 368 /*
 369  * Proportional Set Size(PSS): my share of RSS.
 370  *
 371  * PSS of a process is the count of pages it has in memory, where each
 372  * page is divided by the number of processes sharing it.  So if a
 373  * process has 1000 pages all to itself, and 1000 shared with one other
 374  * process, its PSS will be 1500.
 375  *
 376  * To keep (accumulated) division errors low, we adopt a 64bit
 377  * fixed-point pss counter to minimize division errors. So (pss >>
 378  * PSS_SHIFT) would be the real byte count.
 379  *
 380  * A shift of 12 before division means (assuming 4K page size):
 381  *      - 1M 3-user-pages add up to 8KB errors;
 382  *      - supports mapcount up to 2^24, or 16M;
 383  *      - supports PSS up to 2^52 bytes, or 4PB.
 384  */
 385 #define PSS_SHIFT 12
 386
 387 #ifdef CONFIG_PROC_PAGE_MONITOR
 388 struct mem_size_stats {
 389         unsigned long resident;
 390         unsigned long shared_clean;
 391         unsigned long shared_dirty;
 392         unsigned long private_clean;
 393         unsigned long private_dirty;
 394         unsigned long referenced;
 395         unsigned long anonymous;
 396         unsigned long lazyfree;
 397         unsigned long anonymous_thp;
 398         unsigned long shmem_thp;
 399         unsigned long file_thp;
 400         unsigned long swap;
 401         unsigned long shared_hugetlb;
 402         unsigned long private_hugetlb;
 403         unsigned long ksm;
 404         u64 pss;
 405         u64 pss_anon;
 406         u64 pss_file;
 407         u64 pss_shmem;
 408         u64 pss_dirty;
 409         u64 pss_locked;
 410         u64 swap_pss;
 411 };
 412
 413 static void smaps_page_accumulate(struct mem_size_stats *mss,
 414                 struct page *page, unsigned long size, unsigned long pss,
 415                 bool dirty, bool locked, bool private)
 416 {
 417         mss->pss += pss;
 418
 419         if (PageAnon(page))
 420                 mss->pss_anon += pss;
 421         else if (PageSwapBacked(page))
 422                 mss->pss_shmem += pss;
 423         else
 424                 mss->pss_file += pss;
 425
 426         if (locked)
 427                 mss->pss_locked += pss;
 428
 429         if (dirty || PageDirty(page)) {
 430                 mss->pss_dirty += pss;
 431                 if (private)
 432                         mss->private_dirty += size;
 433                 else
 434                         mss->shared_dirty += size;
 435         } else {
 436                 if (private)
 437                         mss->private_clean += size;
 438                 else
 439                         mss->shared_clean += size;
 440         }
 441 }
 442
 443 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 444                 bool compound, bool young, bool dirty, bool locked,
 445                 bool migration)
 446 {
 447         int i, nr = compound ? compound_nr(page) : 1;
 448         unsigned long size = nr * PAGE_SIZE;
 449
 450         /*
 451          * First accumulate quantities that depend only on |size| and the type
 452          * of the compound page.
 453          */
 454         if (PageAnon(page)) {
 455                 mss->anonymous += size;
 456                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 457                         mss->lazyfree += size;
 458         }
 459
 460         if (PageKsm(page))
 461                 mss->ksm += size;
 462
 463         mss->resident += size;
 464         /* Accumulate the size in pages that have been accessed. */
 465         if (young || page_is_young(page) || PageReferenced(page))
 466                 mss->referenced += size;
 467
 468         /*
 469          * Then accumulate quantities that may depend on sharing, or that may
 470          * differ page-by-page.
 471          *
 472          * page_count(page) == 1 guarantees the page is mapped exactly once.
 473          * If any subpage of the compound page mapped with PTE it would elevate
 474          * page_count().
 475          *
 476          * The page_mapcount() is called to get a snapshot of the mapcount.
 477          * Without holding the page lock this snapshot can be slightly wrong as
 478          * we cannot always read the mapcount atomically.  It is not safe to
 479          * call page_mapcount() even with PTL held if the page is not mapped,
 480          * especially for migration entries.  Treat regular migration entries
 481          * as mapcount == 1.
 482          */
 483         if ((page_count(page) == 1) || migration) {
 484                 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
 485                         locked, true);
 486                 return;
 487         }
 488         for (i = 0; i < nr; i++, page++) {
 489                 int mapcount = page_mapcount(page);
 490                 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
 491                 if (mapcount >= 2)
 492                         pss /= mapcount;
 493                 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
 494                                       mapcount < 2);
 495         }
 496 }
 497
 498 #ifdef CONFIG_SHMEM
 499 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 500                           __always_unused int depth, struct mm_walk *walk)
 501 {
 502         struct mem_size_stats *mss = walk->private;
 503         struct vm_area_struct *vma = walk->vma;
 504
 505         mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping,
 506                                               linear_page_index(vma, addr),
 507                                               linear_page_index(vma, end));
 508
 509         return 0;
 510 }
 511 #else
 512 #define smaps_pte_hole          NULL
 513 #endif /* CONFIG_SHMEM */
 514
 515 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
 516 {
 517 #ifdef CONFIG_SHMEM
 518         if (walk->ops->pte_hole) {
 519                 /* depth is not used */
 520                 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
 521         }
 522 #endif
 523 }
 524
 525 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 526                 struct mm_walk *walk)
 527 {
 528         struct mem_size_stats *mss = walk->private;
 529         struct vm_area_struct *vma = walk->vma;
 530         bool locked = !!(vma->vm_flags & VM_LOCKED);
 531         struct page *page = NULL;
 532         bool migration = false, young = false, dirty = false;
 533         pte_t ptent = ptep_get(pte);
 534
 535         if (pte_present(ptent)) {
 536                 page = vm_normal_page(vma, addr, ptent);
 537                 young = pte_young(ptent);
 538                 dirty = pte_dirty(ptent);
 539         } else if (is_swap_pte(ptent)) {
 540                 swp_entry_t swpent = pte_to_swp_entry(ptent);
 541
 542                 if (!non_swap_entry(swpent)) {
 543                         int mapcount;
 544
 545                         mss->swap += PAGE_SIZE;
 546                         mapcount = swp_swapcount(swpent);
 547                         if (mapcount >= 2) {
 548                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 549
 550                                 do_div(pss_delta, mapcount);
 551                                 mss->swap_pss += pss_delta;
 552                         } else {
 553                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 554                         }
 555                 } else if (is_pfn_swap_entry(swpent)) {
 556                         if (is_migration_entry(swpent))
 557                                 migration = true;
 558                         page = pfn_swap_entry_to_page(swpent);
 559                 }
 560         } else {
 561                 smaps_pte_hole_lookup(addr, walk);
 562                 return;
 563         }
 564
 565         if (!page)
 566                 return;
 567
 568         smaps_account(mss, page, false, young, dirty, locked, migration);
 569 }
 570
 571 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 572 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 573                 struct mm_walk *walk)
 574 {
 575         struct mem_size_stats *mss = walk->private;
 576         struct vm_area_struct *vma = walk->vma;
 577         bool locked = !!(vma->vm_flags & VM_LOCKED);
 578         struct page *page = NULL;
 579         bool migration = false;
 580
 581         if (pmd_present(*pmd)) {
 582                 page = vm_normal_page_pmd(vma, addr, *pmd);
 583         } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
 584                 swp_entry_t entry = pmd_to_swp_entry(*pmd);
 585
 586                 if (is_migration_entry(entry)) {
 587                         migration = true;
 588                         page = pfn_swap_entry_to_page(entry);
 589                 }
 590         }
 591         if (IS_ERR_OR_NULL(page))
 592                 return;
 593         if (PageAnon(page))
 594                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 595         else if (PageSwapBacked(page))
 596                 mss->shmem_thp += HPAGE_PMD_SIZE;
 597         else if (is_zone_device_page(page))
 598                 /* pass */;
 599         else
 600                 mss->file_thp += HPAGE_PMD_SIZE;
 601
 602         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
 603                       locked, migration);
 604 }
 605 #else
 606 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 607                 struct mm_walk *walk)
 608 {
 609 }
 610 #endif
 611
 612 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 613                            struct mm_walk *walk)
 614 {
 615         struct vm_area_struct *vma = walk->vma;
 616         pte_t *pte;
 617         spinlock_t *ptl;
 618
 619         ptl = pmd_trans_huge_lock(pmd, vma);
 620         if (ptl) {
 621                 smaps_pmd_entry(pmd, addr, walk);
 622                 spin_unlock(ptl);
 623                 goto out;
 624         }
 625
 626         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 627         if (!pte) {
 628                 walk->action = ACTION_AGAIN;
 629                 return 0;
 630         }
 631         for (; addr != end; pte++, addr += PAGE_SIZE)
 632                 smaps_pte_entry(pte, addr, walk);
 633         pte_unmap_unlock(pte - 1, ptl);
 634 out:
 635         cond_resched();
 636         return 0;
 637 }
 638
 639 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 640 {
 641         /*
 642          * Don't forget to update Documentation/ on changes.
 643          */
 644         static const char mnemonics[BITS_PER_LONG][2] = {
 645                 /*
 646                  * In case if we meet a flag we don't know about.
 647                  */
 648                 [0 ... (BITS_PER_LONG-1)] = "??",
 649
 650                 [ilog2(VM_READ)]        = "rd",
 651                 [ilog2(VM_WRITE)]       = "wr",
 652                 [ilog2(VM_EXEC)]        = "ex",
 653                 [ilog2(VM_SHARED)]      = "sh",
 654                 [ilog2(VM_MAYREAD)]     = "mr",
 655                 [ilog2(VM_MAYWRITE)]    = "mw",
 656                 [ilog2(VM_MAYEXEC)]     = "me",
 657                 [ilog2(VM_MAYSHARE)]    = "ms",
 658                 [ilog2(VM_GROWSDOWN)]   = "gd",
 659                 [ilog2(VM_PFNMAP)]      = "pf",
 660                 [ilog2(VM_LOCKED)]      = "lo",
 661                 [ilog2(VM_IO)]          = "io",
 662                 [ilog2(VM_SEQ_READ)]    = "sr",
 663                 [ilog2(VM_RAND_READ)]   = "rr",
 664                 [ilog2(VM_DONTCOPY)]    = "dc",
 665                 [ilog2(VM_DONTEXPAND)]  = "de",
 666                 [ilog2(VM_LOCKONFAULT)] = "lf",
 667                 [ilog2(VM_ACCOUNT)]     = "ac",
 668                 [ilog2(VM_NORESERVE)]   = "nr",
 669                 [ilog2(VM_HUGETLB)]     = "ht",
 670                 [ilog2(VM_SYNC)]        = "sf",
 671                 [ilog2(VM_ARCH_1)]      = "ar",
 672                 [ilog2(VM_WIPEONFORK)]  = "wf",
 673                 [ilog2(VM_DONTDUMP)]    = "dd",
 674 #ifdef CONFIG_ARM64_BTI
 675                 [ilog2(VM_ARM64_BTI)]   = "bt",
 676 #endif
 677 #ifdef CONFIG_MEM_SOFT_DIRTY
 678                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 679 #endif
 680                 [ilog2(VM_MIXEDMAP)]    = "mm",
 681                 [ilog2(VM_HUGEPAGE)]    = "hg",
 682                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 683                 [ilog2(VM_MERGEABLE)]   = "mg",
 684                 [ilog2(VM_UFFD_MISSING)]= "um",
 685                 [ilog2(VM_UFFD_WP)]     = "uw",
 686 #ifdef CONFIG_ARM64_MTE
 687                 [ilog2(VM_MTE)]         = "mt",
 688                 [ilog2(VM_MTE_ALLOWED)] = "",
 689 #endif
 690 #ifdef CONFIG_ARCH_HAS_PKEYS
 691                 /* These come out via ProtectionKey: */
 692                 [ilog2(VM_PKEY_BIT0)]   = "",
 693                 [ilog2(VM_PKEY_BIT1)]   = "",
 694                 [ilog2(VM_PKEY_BIT2)]   = "",
 695                 [ilog2(VM_PKEY_BIT3)]   = "",
 696 #if VM_PKEY_BIT4
 697                 [ilog2(VM_PKEY_BIT4)]   = "",
 698 #endif
 699 #endif /* CONFIG_ARCH_HAS_PKEYS */
 700 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
 701                 [ilog2(VM_UFFD_MINOR)]  = "ui",
 702 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
 703 #ifdef CONFIG_X86_USER_SHADOW_STACK
 704                 [ilog2(VM_SHADOW_STACK)] = "ss",
 705 #endif
 706         };
 707         size_t i;
 708
 709         seq_puts(m, "VmFlags: ");
 710         for (i = 0; i < BITS_PER_LONG; i++) {
 711                 if (!mnemonics[i][0])
 712                         continue;
 713                 if (vma->vm_flags & (1UL << i)) {
 714                         seq_putc(m, mnemonics[i][0]);
 715                         seq_putc(m, mnemonics[i][1]);
 716                         seq_putc(m, ' ');
 717                 }
 718         }
 719         seq_putc(m, '\n');
 720 }
 721
 722 #ifdef CONFIG_HUGETLB_PAGE
 723 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 724                                  unsigned long addr, unsigned long end,
 725                                  struct mm_walk *walk)
 726 {
 727         struct mem_size_stats *mss = walk->private;
 728         struct vm_area_struct *vma = walk->vma;
 729         struct page *page = NULL;
 730         pte_t ptent = ptep_get(pte);
 731
 732         if (pte_present(ptent)) {
 733                 page = vm_normal_page(vma, addr, ptent);
 734         } else if (is_swap_pte(ptent)) {
 735                 swp_entry_t swpent = pte_to_swp_entry(ptent);
 736
 737                 if (is_pfn_swap_entry(swpent))
 738                         page = pfn_swap_entry_to_page(swpent);
 739         }
 740         if (page) {
 741                 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
 742                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 743                 else
 744                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 745         }
 746         return 0;
 747 }
 748 #else
 749 #define smaps_hugetlb_range     NULL
 750 #endif /* HUGETLB_PAGE */
 751
 752 static const struct mm_walk_ops smaps_walk_ops = {
 753         .pmd_entry              = smaps_pte_range,
 754         .hugetlb_entry          = smaps_hugetlb_range,
 755         .walk_lock              = PGWALK_RDLOCK,
 756 };
 757
 758 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 759         .pmd_entry              = smaps_pte_range,
 760         .hugetlb_entry          = smaps_hugetlb_range,
 761         .pte_hole               = smaps_pte_hole,
 762         .walk_lock              = PGWALK_RDLOCK,
 763 };
 764
 765 /*
 766  * Gather mem stats from @vma with the indicated beginning
 767  * address @start, and keep them in @mss.
 768  *
 769  * Use vm_start of @vma as the beginning address if @start is 0.
 770  */
 771 static void smap_gather_stats(struct vm_area_struct *vma,
 772                 struct mem_size_stats *mss, unsigned long start)
 773 {
 774         const struct mm_walk_ops *ops = &smaps_walk_ops;
 775
 776         /* Invalid start */
 777         if (start >= vma->vm_end)
 778                 return;
 779
 780         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 781                 /*
 782                  * For shared or readonly shmem mappings we know that all
 783                  * swapped out pages belong to the shmem object, and we can
 784                  * obtain the swap value much more efficiently. For private
 785                  * writable mappings, we might have COW pages that are
 786                  * not affected by the parent swapped out pages of the shmem
 787                  * object, so we have to distinguish them during the page walk.
 788                  * Unless we know that the shmem object (or the part mapped by
 789                  * our VMA) has no swapped out pages at all.
 790                  */
 791                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 792
 793                 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 794                                         !(vma->vm_flags & VM_WRITE))) {
 795                         mss->swap += shmem_swapped;
 796                 } else {
 797                         ops = &smaps_shmem_walk_ops;
 798                 }
 799         }
 800
 801         /* mmap_lock is held in m_start */
 802         if (!start)
 803                 walk_page_vma(vma, ops, mss);
 804         else
 805                 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
 806 }
 807
 808 #define SEQ_PUT_DEC(str, val) \
 809                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 810
 811 /* Show the contents common for smaps and smaps_rollup */
 812 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 813         bool rollup_mode)
 814 {
 815         SEQ_PUT_DEC("Rss:            ", mss->resident);
 816         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 817         SEQ_PUT_DEC(" kB\nPss_Dirty:      ", mss->pss_dirty >> PSS_SHIFT);
 818         if (rollup_mode) {
 819                 /*
 820                  * These are meaningful only for smaps_rollup, otherwise two of
 821                  * them are zero, and the other one is the same as Pss.
 822                  */
 823                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 824                         mss->pss_anon >> PSS_SHIFT);
 825                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 826                         mss->pss_file >> PSS_SHIFT);
 827                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 828                         mss->pss_shmem >> PSS_SHIFT);
 829         }
 830         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 831         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 832         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 833         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 834         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 835         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 836         SEQ_PUT_DEC(" kB\nKSM:            ", mss->ksm);
 837         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 838         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 839         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 840         SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
 841         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 842         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 843                                   mss->private_hugetlb >> 10, 7);
 844         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 845         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 846                                         mss->swap_pss >> PSS_SHIFT);
 847         SEQ_PUT_DEC(" kB\nLocked:         ",
 848                                         mss->pss_locked >> PSS_SHIFT);
 849         seq_puts(m, " kB\n");
 850 }
 851
 852 static int show_smap(struct seq_file *m, void *v)
 853 {
 854         struct vm_area_struct *vma = v;
 855         struct mem_size_stats mss = {};
 856
 857         smap_gather_stats(vma, &mss, 0);
 858
 859         show_map_vma(m, vma);
 860
 861         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 862         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 863         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 864         seq_puts(m, " kB\n");
 865
 866         __show_smap(m, &mss, false);
 867
 868         seq_printf(m, "THPeligible:    %8u\n",
 869                    !!thp_vma_allowable_orders(vma, vma->vm_flags, true, false,
 870                                               true, THP_ORDERS_ALL));
 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         ret = mmap_read_lock_killable(mm);
 899         if (ret)
 900                 goto out_put_mm;
 901
 902         hold_task_mempolicy(priv);
 903         vma = vma_next(&vmi);
 904
 905         if (unlikely(!vma))
 906                 goto empty_set;
 907
 908         vma_start = vma->vm_start;
 909         do {
 910                 smap_gather_stats(vma, &mss, 0);
 911                 last_vma_end = vma->vm_end;
 912
 913                 /*
 914                  * Release mmap_lock temporarily if someone wants to
 915                  * access it for write request.
 916                  */
 917                 if (mmap_lock_is_contended(mm)) {
 918                         vma_iter_invalidate(&vmi);
 919                         mmap_read_unlock(mm);
 920                         ret = mmap_read_lock_killable(mm);
 921                         if (ret) {
 922                                 release_task_mempolicy(priv);
 923                                 goto out_put_mm;
 924                         }
 925
 926                         /*
 927                          * After dropping the lock, there are four cases to
 928                          * consider. See the following example for explanation.
 929                          *
 930                          *   +------+------+-----------+
 931                          *   | VMA1 | VMA2 | VMA3      |
 932                          *   +------+------+-----------+
 933                          *   |      |      |           |
 934                          *  4k     8k     16k         400k
 935                          *
 936                          * Suppose we drop the lock after reading VMA2 due to
 937                          * contention, then we get:
 938                          *
 939                          *      last_vma_end = 16k
 940                          *
 941                          * 1) VMA2 is freed, but VMA3 exists:
 942                          *
 943                          *    vma_next(vmi) will return VMA3.
 944                          *    In this case, just continue from VMA3.
 945                          *
 946                          * 2) VMA2 still exists:
 947                          *
 948                          *    vma_next(vmi) will return VMA3.
 949                          *    In this case, just continue from VMA3.
 950                          *
 951                          * 3) No more VMAs can be found:
 952                          *
 953                          *    vma_next(vmi) will return NULL.
 954                          *    No more things to do, just break.
 955                          *
 956                          * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
 957                          *
 958                          *    vma_next(vmi) will return VMA' whose range
 959                          *    contains last_vma_end.
 960                          *    Iterate VMA' from last_vma_end.
 961                          */
 962                         vma = vma_next(&vmi);
 963                         /* Case 3 above */
 964                         if (!vma)
 965                                 break;
 966
 967                         /* Case 1 and 2 above */
 968                         if (vma->vm_start >= last_vma_end)
 969                                 continue;
 970
 971                         /* Case 4 above */
 972                         if (vma->vm_end > last_vma_end)
 973                                 smap_gather_stats(vma, &mss, last_vma_end);
 974                 }
 975         } for_each_vma(vmi, vma);
 976
 977 empty_set:
 978         show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
 979         seq_pad(m, ' ');
 980         seq_puts(m, "[rollup]\n");
 981
 982         __show_smap(m, &mss, true);
 983
 984         release_task_mempolicy(priv);
 985         mmap_read_unlock(mm);
 986
 987 out_put_mm:
 988         mmput(mm);
 989 out_put_task:
 990         put_task_struct(priv->task);
 991         priv->task = NULL;
 992
 993         return ret;
 994 }
 995 #undef SEQ_PUT_DEC
 996
 997 static const struct seq_operations proc_pid_smaps_op = {
 998         .start  = m_start,
 999         .next   = m_next,
1000         .stop   = m_stop,
1001         .show   = show_smap
1002 };
1003
1004 static int pid_smaps_open(struct inode *inode, struct file *file)
1005 {
1006         return do_maps_open(inode, file, &proc_pid_smaps_op);
1007 }
1008
1009 static int smaps_rollup_open(struct inode *inode, struct file *file)
1010 {
1011         int ret;
1012         struct proc_maps_private *priv;
1013
1014         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1015         if (!priv)
1016                 return -ENOMEM;
1017
1018         ret = single_open(file, show_smaps_rollup, priv);
1019         if (ret)
1020                 goto out_free;
1021
1022         priv->inode = inode;
1023         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1024         if (IS_ERR(priv->mm)) {
1025                 ret = PTR_ERR(priv->mm);
1026
1027                 single_release(inode, file);
1028                 goto out_free;
1029         }
1030
1031         return 0;
1032
1033 out_free:
1034         kfree(priv);
1035         return ret;
1036 }
1037
1038 static int smaps_rollup_release(struct inode *inode, struct file *file)
1039 {
1040         struct seq_file *seq = file->private_data;
1041         struct proc_maps_private *priv = seq->private;
1042
1043         if (priv->mm)
1044                 mmdrop(priv->mm);
1045
1046         kfree(priv);
1047         return single_release(inode, file);
1048 }
1049
1050 const struct file_operations proc_pid_smaps_operations = {
1051         .open           = pid_smaps_open,
1052         .read           = seq_read,
1053         .llseek         = seq_lseek,
1054         .release        = proc_map_release,
1055 };
1056
1057 const struct file_operations proc_pid_smaps_rollup_operations = {
1058         .open           = smaps_rollup_open,
1059         .read           = seq_read,
1060         .llseek         = seq_lseek,
1061         .release        = smaps_rollup_release,
1062 };
1063
1064 enum clear_refs_types {
1065         CLEAR_REFS_ALL = 1,
1066         CLEAR_REFS_ANON,
1067         CLEAR_REFS_MAPPED,
1068         CLEAR_REFS_SOFT_DIRTY,
1069         CLEAR_REFS_MM_HIWATER_RSS,
1070         CLEAR_REFS_LAST,
1071 };
1072
1073 struct clear_refs_private {
1074         enum clear_refs_types type;
1075 };
1076
1077 #ifdef CONFIG_MEM_SOFT_DIRTY
1078
1079 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1080 {
1081         struct page *page;
1082
1083         if (!pte_write(pte))
1084                 return false;
1085         if (!is_cow_mapping(vma->vm_flags))
1086                 return false;
1087         if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1088                 return false;
1089         page = vm_normal_page(vma, addr, pte);
1090         if (!page)
1091                 return false;
1092         return page_maybe_dma_pinned(page);
1093 }
1094
1095 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1096                 unsigned long addr, pte_t *pte)
1097 {
1098         /*
1099          * The soft-dirty tracker uses #PF-s to catch writes
1100          * to pages, so write-protect the pte as well. See the
1101          * Documentation/admin-guide/mm/soft-dirty.rst for full description
1102          * of how soft-dirty works.
1103          */
1104         pte_t ptent = ptep_get(pte);
1105
1106         if (pte_present(ptent)) {
1107                 pte_t old_pte;
1108
1109                 if (pte_is_pinned(vma, addr, ptent))
1110                         return;
1111                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1112                 ptent = pte_wrprotect(old_pte);
1113                 ptent = pte_clear_soft_dirty(ptent);
1114                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1115         } else if (is_swap_pte(ptent)) {
1116                 ptent = pte_swp_clear_soft_dirty(ptent);
1117                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1118         }
1119 }
1120 #else
1121 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1122                 unsigned long addr, pte_t *pte)
1123 {
1124 }
1125 #endif
1126
1127 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1128 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1129                 unsigned long addr, pmd_t *pmdp)
1130 {
1131         pmd_t old, pmd = *pmdp;
1132
1133         if (pmd_present(pmd)) {
1134                 /* See comment in change_huge_pmd() */
1135                 old = pmdp_invalidate(vma, addr, pmdp);
1136                 if (pmd_dirty(old))
1137                         pmd = pmd_mkdirty(pmd);
1138                 if (pmd_young(old))
1139                         pmd = pmd_mkyoung(pmd);
1140
1141                 pmd = pmd_wrprotect(pmd);
1142                 pmd = pmd_clear_soft_dirty(pmd);
1143
1144                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1145         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1146                 pmd = pmd_swp_clear_soft_dirty(pmd);
1147                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1148         }
1149 }
1150 #else
1151 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1152                 unsigned long addr, pmd_t *pmdp)
1153 {
1154 }
1155 #endif
1156
1157 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1158                                 unsigned long end, struct mm_walk *walk)
1159 {
1160         struct clear_refs_private *cp = walk->private;
1161         struct vm_area_struct *vma = walk->vma;
1162         pte_t *pte, ptent;
1163         spinlock_t *ptl;
1164         struct page *page;
1165
1166         ptl = pmd_trans_huge_lock(pmd, vma);
1167         if (ptl) {
1168                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1169                         clear_soft_dirty_pmd(vma, addr, pmd);
1170                         goto out;
1171                 }
1172
1173                 if (!pmd_present(*pmd))
1174                         goto out;
1175
1176                 page = pmd_page(*pmd);
1177
1178                 /* Clear accessed and referenced bits. */
1179                 pmdp_test_and_clear_young(vma, addr, pmd);
1180                 test_and_clear_page_young(page);
1181                 ClearPageReferenced(page);
1182 out:
1183                 spin_unlock(ptl);
1184                 return 0;
1185         }
1186
1187         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1188         if (!pte) {
1189                 walk->action = ACTION_AGAIN;
1190                 return 0;
1191         }
1192         for (; addr != end; pte++, addr += PAGE_SIZE) {
1193                 ptent = ptep_get(pte);
1194
1195                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1196                         clear_soft_dirty(vma, addr, pte);
1197                         continue;
1198                 }
1199
1200                 if (!pte_present(ptent))
1201                         continue;
1202
1203                 page = vm_normal_page(vma, addr, ptent);
1204                 if (!page)
1205                         continue;
1206
1207                 /* Clear accessed and referenced bits. */
1208                 ptep_test_and_clear_young(vma, addr, pte);
1209                 test_and_clear_page_young(page);
1210                 ClearPageReferenced(page);
1211         }
1212         pte_unmap_unlock(pte - 1, ptl);
1213         cond_resched();
1214         return 0;
1215 }
1216
1217 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1218                                 struct mm_walk *walk)
1219 {
1220         struct clear_refs_private *cp = walk->private;
1221         struct vm_area_struct *vma = walk->vma;
1222
1223         if (vma->vm_flags & VM_PFNMAP)
1224                 return 1;
1225
1226         /*
1227          * Writing 1 to /proc/pid/clear_refs affects all pages.
1228          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1229          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1230          * Writing 4 to /proc/pid/clear_refs affects all pages.
1231          */
1232         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1233                 return 1;
1234         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1235                 return 1;
1236         return 0;
1237 }
1238
1239 static const struct mm_walk_ops clear_refs_walk_ops = {
1240         .pmd_entry              = clear_refs_pte_range,
1241         .test_walk              = clear_refs_test_walk,
1242         .walk_lock              = PGWALK_WRLOCK,
1243 };
1244
1245 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1246                                 size_t count, loff_t *ppos)
1247 {
1248         struct task_struct *task;
1249         char buffer[PROC_NUMBUF] = {};
1250         struct mm_struct *mm;
1251         struct vm_area_struct *vma;
1252         enum clear_refs_types type;
1253         int itype;
1254         int rv;
1255
1256         if (count > sizeof(buffer) - 1)
1257                 count = sizeof(buffer) - 1;
1258         if (copy_from_user(buffer, buf, count))
1259                 return -EFAULT;
1260         rv = kstrtoint(strstrip(buffer), 10, &itype);
1261         if (rv < 0)
1262                 return rv;
1263         type = (enum clear_refs_types)itype;
1264         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1265                 return -EINVAL;
1266
1267         task = get_proc_task(file_inode(file));
1268         if (!task)
1269                 return -ESRCH;
1270         mm = get_task_mm(task);
1271         if (mm) {
1272                 VMA_ITERATOR(vmi, mm, 0);
1273                 struct mmu_notifier_range range;
1274                 struct clear_refs_private cp = {
1275                         .type = type,
1276                 };
1277
1278                 if (mmap_write_lock_killable(mm)) {
1279                         count = -EINTR;
1280                         goto out_mm;
1281                 }
1282                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1283                         /*
1284                          * Writing 5 to /proc/pid/clear_refs resets the peak
1285                          * resident set size to this mm's current rss value.
1286                          */
1287                         reset_mm_hiwater_rss(mm);
1288                         goto out_unlock;
1289                 }
1290
1291                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1292                         for_each_vma(vmi, vma) {
1293                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1294                                         continue;
1295                                 vm_flags_clear(vma, VM_SOFTDIRTY);
1296                                 vma_set_page_prot(vma);
1297                         }
1298
1299                         inc_tlb_flush_pending(mm);
1300                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1301                                                 0, mm, 0, -1UL);
1302                         mmu_notifier_invalidate_range_start(&range);
1303                 }
1304                 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1305                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1306                         mmu_notifier_invalidate_range_end(&range);
1307                         flush_tlb_mm(mm);
1308                         dec_tlb_flush_pending(mm);
1309                 }
1310 out_unlock:
1311                 mmap_write_unlock(mm);
1312 out_mm:
1313                 mmput(mm);
1314         }
1315         put_task_struct(task);
1316
1317         return count;
1318 }
1319
1320 const struct file_operations proc_clear_refs_operations = {
1321         .write          = clear_refs_write,
1322         .llseek         = noop_llseek,
1323 };
1324
1325 typedef struct {
1326         u64 pme;
1327 } pagemap_entry_t;
1328
1329 struct pagemapread {
1330         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1331         pagemap_entry_t *buffer;
1332         bool show_pfn;
1333 };
1334
1335 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1336 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1337
1338 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1339 #define PM_PFRAME_BITS          55
1340 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1341 #define PM_SOFT_DIRTY           BIT_ULL(55)
1342 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1343 #define PM_UFFD_WP              BIT_ULL(57)
1344 #define PM_FILE                 BIT_ULL(61)
1345 #define PM_SWAP                 BIT_ULL(62)
1346 #define PM_PRESENT              BIT_ULL(63)
1347
1348 #define PM_END_OF_BUFFER    1
1349
1350 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1351 {
1352         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1353 }
1354
1355 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1356                           struct pagemapread *pm)
1357 {
1358         pm->buffer[pm->pos++] = *pme;
1359         if (pm->pos >= pm->len)
1360                 return PM_END_OF_BUFFER;
1361         return 0;
1362 }
1363
1364 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1365                             __always_unused int depth, struct mm_walk *walk)
1366 {
1367         struct pagemapread *pm = walk->private;
1368         unsigned long addr = start;
1369         int err = 0;
1370
1371         while (addr < end) {
1372                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1373                 pagemap_entry_t pme = make_pme(0, 0);
1374                 /* End of address space hole, which we mark as non-present. */
1375                 unsigned long hole_end;
1376
1377                 if (vma)
1378                         hole_end = min(end, vma->vm_start);
1379                 else
1380                         hole_end = end;
1381
1382                 for (; addr < hole_end; addr += PAGE_SIZE) {
1383                         err = add_to_pagemap(addr, &pme, pm);
1384                         if (err)
1385                                 goto out;
1386                 }
1387
1388                 if (!vma)
1389                         break;
1390
1391                 /* Addresses in the VMA. */
1392                 if (vma->vm_flags & VM_SOFTDIRTY)
1393                         pme = make_pme(0, PM_SOFT_DIRTY);
1394                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1395                         err = add_to_pagemap(addr, &pme, pm);
1396                         if (err)
1397                                 goto out;
1398                 }
1399         }
1400 out:
1401         return err;
1402 }
1403
1404 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1405                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1406 {
1407         u64 frame = 0, flags = 0;
1408         struct page *page = NULL;
1409         bool migration = false;
1410
1411         if (pte_present(pte)) {
1412                 if (pm->show_pfn)
1413                         frame = pte_pfn(pte);
1414                 flags |= PM_PRESENT;
1415                 page = vm_normal_page(vma, addr, pte);
1416                 if (pte_soft_dirty(pte))
1417                         flags |= PM_SOFT_DIRTY;
1418                 if (pte_uffd_wp(pte))
1419                         flags |= PM_UFFD_WP;
1420         } else if (is_swap_pte(pte)) {
1421                 swp_entry_t entry;
1422                 if (pte_swp_soft_dirty(pte))
1423                         flags |= PM_SOFT_DIRTY;
1424                 if (pte_swp_uffd_wp(pte))
1425                         flags |= PM_UFFD_WP;
1426                 entry = pte_to_swp_entry(pte);
1427                 if (pm->show_pfn) {
1428                         pgoff_t offset;
1429                         /*
1430                          * For PFN swap offsets, keeping the offset field
1431                          * to be PFN only to be compatible with old smaps.
1432                          */
1433                         if (is_pfn_swap_entry(entry))
1434                                 offset = swp_offset_pfn(entry);
1435                         else
1436                                 offset = swp_offset(entry);
1437                         frame = swp_type(entry) |
1438                             (offset << MAX_SWAPFILES_SHIFT);
1439                 }
1440                 flags |= PM_SWAP;
1441                 migration = is_migration_entry(entry);
1442                 if (is_pfn_swap_entry(entry))
1443                         page = pfn_swap_entry_to_page(entry);
1444                 if (pte_marker_entry_uffd_wp(entry))
1445                         flags |= PM_UFFD_WP;
1446         }
1447
1448         if (page && !PageAnon(page))
1449                 flags |= PM_FILE;
1450         if (page && !migration && page_mapcount(page) == 1)
1451                 flags |= PM_MMAP_EXCLUSIVE;
1452         if (vma->vm_flags & VM_SOFTDIRTY)
1453                 flags |= PM_SOFT_DIRTY;
1454
1455         return make_pme(frame, flags);
1456 }
1457
1458 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1459                              struct mm_walk *walk)
1460 {
1461         struct vm_area_struct *vma = walk->vma;
1462         struct pagemapread *pm = walk->private;
1463         spinlock_t *ptl;
1464         pte_t *pte, *orig_pte;
1465         int err = 0;
1466 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1467         bool migration = false;
1468
1469         ptl = pmd_trans_huge_lock(pmdp, vma);
1470         if (ptl) {
1471                 u64 flags = 0, frame = 0;
1472                 pmd_t pmd = *pmdp;
1473                 struct page *page = NULL;
1474
1475                 if (vma->vm_flags & VM_SOFTDIRTY)
1476                         flags |= PM_SOFT_DIRTY;
1477
1478                 if (pmd_present(pmd)) {
1479                         page = pmd_page(pmd);
1480
1481                         flags |= PM_PRESENT;
1482                         if (pmd_soft_dirty(pmd))
1483                                 flags |= PM_SOFT_DIRTY;
1484                         if (pmd_uffd_wp(pmd))
1485                                 flags |= PM_UFFD_WP;
1486                         if (pm->show_pfn)
1487                                 frame = pmd_pfn(pmd) +
1488                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1489                 }
1490 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1491                 else if (is_swap_pmd(pmd)) {
1492                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1493                         unsigned long offset;
1494
1495                         if (pm->show_pfn) {
1496                                 if (is_pfn_swap_entry(entry))
1497                                         offset = swp_offset_pfn(entry);
1498                                 else
1499                                         offset = swp_offset(entry);
1500                                 offset = offset +
1501                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1502                                 frame = swp_type(entry) |
1503                                         (offset << MAX_SWAPFILES_SHIFT);
1504                         }
1505                         flags |= PM_SWAP;
1506                         if (pmd_swp_soft_dirty(pmd))
1507                                 flags |= PM_SOFT_DIRTY;
1508                         if (pmd_swp_uffd_wp(pmd))
1509                                 flags |= PM_UFFD_WP;
1510                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1511                         migration = is_migration_entry(entry);
1512                         page = pfn_swap_entry_to_page(entry);
1513                 }
1514 #endif
1515
1516                 if (page && !migration && page_mapcount(page) == 1)
1517                         flags |= PM_MMAP_EXCLUSIVE;
1518
1519                 for (; addr != end; addr += PAGE_SIZE) {
1520                         pagemap_entry_t pme = make_pme(frame, flags);
1521
1522                         err = add_to_pagemap(addr, &pme, pm);
1523                         if (err)
1524                                 break;
1525                         if (pm->show_pfn) {
1526                                 if (flags & PM_PRESENT)
1527                                         frame++;
1528                                 else if (flags & PM_SWAP)
1529                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1530                         }
1531                 }
1532                 spin_unlock(ptl);
1533                 return err;
1534         }
1535 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1536
1537         /*
1538          * We can assume that @vma always points to a valid one and @end never
1539          * goes beyond vma->vm_end.
1540          */
1541         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1542         if (!pte) {
1543                 walk->action = ACTION_AGAIN;
1544                 return err;
1545         }
1546         for (; addr < end; pte++, addr += PAGE_SIZE) {
1547                 pagemap_entry_t pme;
1548
1549                 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte));
1550                 err = add_to_pagemap(addr, &pme, pm);
1551                 if (err)
1552                         break;
1553         }
1554         pte_unmap_unlock(orig_pte, ptl);
1555
1556         cond_resched();
1557
1558         return err;
1559 }
1560
1561 #ifdef CONFIG_HUGETLB_PAGE
1562 /* This function walks within one hugetlb entry in the single call */
1563 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1564                                  unsigned long addr, unsigned long end,
1565                                  struct mm_walk *walk)
1566 {
1567         struct pagemapread *pm = walk->private;
1568         struct vm_area_struct *vma = walk->vma;
1569         u64 flags = 0, frame = 0;
1570         int err = 0;
1571         pte_t pte;
1572
1573         if (vma->vm_flags & VM_SOFTDIRTY)
1574                 flags |= PM_SOFT_DIRTY;
1575
1576         pte = huge_ptep_get(ptep);
1577         if (pte_present(pte)) {
1578                 struct page *page = pte_page(pte);
1579
1580                 if (!PageAnon(page))
1581                         flags |= PM_FILE;
1582
1583                 if (page_mapcount(page) == 1)
1584                         flags |= PM_MMAP_EXCLUSIVE;
1585
1586                 if (huge_pte_uffd_wp(pte))
1587                         flags |= PM_UFFD_WP;
1588
1589                 flags |= PM_PRESENT;
1590                 if (pm->show_pfn)
1591                         frame = pte_pfn(pte) +
1592                                 ((addr & ~hmask) >> PAGE_SHIFT);
1593         } else if (pte_swp_uffd_wp_any(pte)) {
1594                 flags |= PM_UFFD_WP;
1595         }
1596
1597         for (; addr != end; addr += PAGE_SIZE) {
1598                 pagemap_entry_t pme = make_pme(frame, flags);
1599
1600                 err = add_to_pagemap(addr, &pme, pm);
1601                 if (err)
1602                         return err;
1603                 if (pm->show_pfn && (flags & PM_PRESENT))
1604                         frame++;
1605         }
1606
1607         cond_resched();
1608
1609         return err;
1610 }
1611 #else
1612 #define pagemap_hugetlb_range   NULL
1613 #endif /* HUGETLB_PAGE */
1614
1615 static const struct mm_walk_ops pagemap_ops = {
1616         .pmd_entry      = pagemap_pmd_range,
1617         .pte_hole       = pagemap_pte_hole,
1618         .hugetlb_entry  = pagemap_hugetlb_range,
1619         .walk_lock      = PGWALK_RDLOCK,
1620 };
1621
1622 /*
1623  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1624  *
1625  * For each page in the address space, this file contains one 64-bit entry
1626  * consisting of the following:
1627  *
1628  * Bits 0-54  page frame number (PFN) if present
1629  * Bits 0-4   swap type if swapped
1630  * Bits 5-54  swap offset if swapped
1631  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1632  * Bit  56    page exclusively mapped
1633  * Bit  57    pte is uffd-wp write-protected
1634  * Bits 58-60 zero
1635  * Bit  61    page is file-page or shared-anon
1636  * Bit  62    page swapped
1637  * Bit  63    page present
1638  *
1639  * If the page is not present but in swap, then the PFN contains an
1640  * encoding of the swap file number and the page's offset into the
1641  * swap. Unmapped pages return a null PFN. This allows determining
1642  * precisely which pages are mapped (or in swap) and comparing mapped
1643  * pages between processes.
1644  *
1645  * Efficient users of this interface will use /proc/pid/maps to
1646  * determine which areas of memory are actually mapped and llseek to
1647  * skip over unmapped regions.
1648  */
1649 static ssize_t pagemap_read(struct file *file, char __user *buf,
1650                             size_t count, loff_t *ppos)
1651 {
1652         struct mm_struct *mm = file->private_data;
1653         struct pagemapread pm;
1654         unsigned long src;
1655         unsigned long svpfn;
1656         unsigned long start_vaddr;
1657         unsigned long end_vaddr;
1658         int ret = 0, copied = 0;
1659
1660         if (!mm || !mmget_not_zero(mm))
1661                 goto out;
1662
1663         ret = -EINVAL;
1664         /* file position must be aligned */
1665         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1666                 goto out_mm;
1667
1668         ret = 0;
1669         if (!count)
1670                 goto out_mm;
1671
1672         /* do not disclose physical addresses: attack vector */
1673         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1674
1675         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1676         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1677         ret = -ENOMEM;
1678         if (!pm.buffer)
1679                 goto out_mm;
1680
1681         src = *ppos;
1682         svpfn = src / PM_ENTRY_BYTES;
1683         end_vaddr = mm->task_size;
1684
1685         /* watch out for wraparound */
1686         start_vaddr = end_vaddr;
1687         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
1688                 unsigned long end;
1689
1690                 ret = mmap_read_lock_killable(mm);
1691                 if (ret)
1692                         goto out_free;
1693                 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
1694                 mmap_read_unlock(mm);
1695
1696                 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
1697                 if (end >= start_vaddr && end < mm->task_size)
1698                         end_vaddr = end;
1699         }
1700
1701         /* Ensure the address is inside the task */
1702         if (start_vaddr > mm->task_size)
1703                 start_vaddr = end_vaddr;
1704
1705         ret = 0;
1706         while (count && (start_vaddr < end_vaddr)) {
1707                 int len;
1708                 unsigned long end;
1709
1710                 pm.pos = 0;
1711                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1712                 /* overflow ? */
1713                 if (end < start_vaddr || end > end_vaddr)
1714                         end = end_vaddr;
1715                 ret = mmap_read_lock_killable(mm);
1716                 if (ret)
1717                         goto out_free;
1718                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1719                 mmap_read_unlock(mm);
1720                 start_vaddr = end;
1721
1722                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1723                 if (copy_to_user(buf, pm.buffer, len)) {
1724                         ret = -EFAULT;
1725                         goto out_free;
1726                 }
1727                 copied += len;
1728                 buf += len;
1729                 count -= len;
1730         }
1731         *ppos += copied;
1732         if (!ret || ret == PM_END_OF_BUFFER)
1733                 ret = copied;
1734
1735 out_free:
1736         kfree(pm.buffer);
1737 out_mm:
1738         mmput(mm);
1739 out:
1740         return ret;
1741 }
1742
1743 static int pagemap_open(struct inode *inode, struct file *file)
1744 {
1745         struct mm_struct *mm;
1746
1747         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1748         if (IS_ERR(mm))
1749                 return PTR_ERR(mm);
1750         file->private_data = mm;
1751         return 0;
1752 }
1753
1754 static int pagemap_release(struct inode *inode, struct file *file)
1755 {
1756         struct mm_struct *mm = file->private_data;
1757
1758         if (mm)
1759                 mmdrop(mm);
1760         return 0;
1761 }
1762
1763 #define PM_SCAN_CATEGORIES      (PAGE_IS_WPALLOWED | PAGE_IS_WRITTEN |  \
1764                                  PAGE_IS_FILE | PAGE_IS_PRESENT |       \
1765                                  PAGE_IS_SWAPPED | PAGE_IS_PFNZERO |    \
1766                                  PAGE_IS_HUGE | PAGE_IS_SOFT_DIRTY)
1767 #define PM_SCAN_FLAGS           (PM_SCAN_WP_MATCHING | PM_SCAN_CHECK_WPASYNC)
1768
1769 struct pagemap_scan_private {
1770         struct pm_scan_arg arg;
1771         unsigned long masks_of_interest, cur_vma_category;
1772         struct page_region *vec_buf;
1773         unsigned long vec_buf_len, vec_buf_index, found_pages;
1774         struct page_region __user *vec_out;
1775 };
1776
1777 static unsigned long pagemap_page_category(struct pagemap_scan_private *p,
1778                                            struct vm_area_struct *vma,
1779                                            unsigned long addr, pte_t pte)
1780 {
1781         unsigned long categories = 0;
1782
1783         if (pte_present(pte)) {
1784                 struct page *page;
1785
1786                 categories |= PAGE_IS_PRESENT;
1787                 if (!pte_uffd_wp(pte))
1788                         categories |= PAGE_IS_WRITTEN;
1789
1790                 if (p->masks_of_interest & PAGE_IS_FILE) {
1791                         page = vm_normal_page(vma, addr, pte);
1792                         if (page && !PageAnon(page))
1793                                 categories |= PAGE_IS_FILE;
1794                 }
1795
1796                 if (is_zero_pfn(pte_pfn(pte)))
1797                         categories |= PAGE_IS_PFNZERO;
1798                 if (pte_soft_dirty(pte))
1799                         categories |= PAGE_IS_SOFT_DIRTY;
1800         } else if (is_swap_pte(pte)) {
1801                 swp_entry_t swp;
1802
1803                 categories |= PAGE_IS_SWAPPED;
1804                 if (!pte_swp_uffd_wp_any(pte))
1805                         categories |= PAGE_IS_WRITTEN;
1806
1807                 if (p->masks_of_interest & PAGE_IS_FILE) {
1808                         swp = pte_to_swp_entry(pte);
1809                         if (is_pfn_swap_entry(swp) &&
1810                             !PageAnon(pfn_swap_entry_to_page(swp)))
1811                                 categories |= PAGE_IS_FILE;
1812                 }
1813                 if (pte_swp_soft_dirty(pte))
1814                         categories |= PAGE_IS_SOFT_DIRTY;
1815         }
1816
1817         return categories;
1818 }
1819
1820 static void make_uffd_wp_pte(struct vm_area_struct *vma,
1821                              unsigned long addr, pte_t *pte)
1822 {
1823         pte_t ptent = ptep_get(pte);
1824
1825         if (pte_present(ptent)) {
1826                 pte_t old_pte;
1827
1828                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1829                 ptent = pte_mkuffd_wp(ptent);
1830                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1831         } else if (is_swap_pte(ptent)) {
1832                 ptent = pte_swp_mkuffd_wp(ptent);
1833                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1834         } else {
1835                 set_pte_at(vma->vm_mm, addr, pte,
1836                            make_pte_marker(PTE_MARKER_UFFD_WP));
1837         }
1838 }
1839
1840 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1841 static unsigned long pagemap_thp_category(struct pagemap_scan_private *p,
1842                                           struct vm_area_struct *vma,
1843                                           unsigned long addr, pmd_t pmd)
1844 {
1845         unsigned long categories = PAGE_IS_HUGE;
1846
1847         if (pmd_present(pmd)) {
1848                 struct page *page;
1849
1850                 categories |= PAGE_IS_PRESENT;
1851                 if (!pmd_uffd_wp(pmd))
1852                         categories |= PAGE_IS_WRITTEN;
1853
1854                 if (p->masks_of_interest & PAGE_IS_FILE) {
1855                         page = vm_normal_page_pmd(vma, addr, pmd);
1856                         if (page && !PageAnon(page))
1857                                 categories |= PAGE_IS_FILE;
1858                 }
1859
1860                 if (is_zero_pfn(pmd_pfn(pmd)))
1861                         categories |= PAGE_IS_PFNZERO;
1862                 if (pmd_soft_dirty(pmd))
1863                         categories |= PAGE_IS_SOFT_DIRTY;
1864         } else if (is_swap_pmd(pmd)) {
1865                 swp_entry_t swp;
1866
1867                 categories |= PAGE_IS_SWAPPED;
1868                 if (!pmd_swp_uffd_wp(pmd))
1869                         categories |= PAGE_IS_WRITTEN;
1870                 if (pmd_swp_soft_dirty(pmd))
1871                         categories |= PAGE_IS_SOFT_DIRTY;
1872
1873                 if (p->masks_of_interest & PAGE_IS_FILE) {
1874                         swp = pmd_to_swp_entry(pmd);
1875                         if (is_pfn_swap_entry(swp) &&
1876                             !PageAnon(pfn_swap_entry_to_page(swp)))
1877                                 categories |= PAGE_IS_FILE;
1878                 }
1879         }
1880
1881         return categories;
1882 }
1883
1884 static void make_uffd_wp_pmd(struct vm_area_struct *vma,
1885                              unsigned long addr, pmd_t *pmdp)
1886 {
1887         pmd_t old, pmd = *pmdp;
1888
1889         if (pmd_present(pmd)) {
1890                 old = pmdp_invalidate_ad(vma, addr, pmdp);
1891                 pmd = pmd_mkuffd_wp(old);
1892                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1893         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1894                 pmd = pmd_swp_mkuffd_wp(pmd);
1895                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1896         }
1897 }
1898 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1899
1900 #ifdef CONFIG_HUGETLB_PAGE
1901 static unsigned long pagemap_hugetlb_category(pte_t pte)
1902 {
1903         unsigned long categories = PAGE_IS_HUGE;
1904
1905         /*
1906          * According to pagemap_hugetlb_range(), file-backed HugeTLB
1907          * page cannot be swapped. So PAGE_IS_FILE is not checked for
1908          * swapped pages.
1909          */
1910         if (pte_present(pte)) {
1911                 categories |= PAGE_IS_PRESENT;
1912                 if (!huge_pte_uffd_wp(pte))
1913                         categories |= PAGE_IS_WRITTEN;
1914                 if (!PageAnon(pte_page(pte)))
1915                         categories |= PAGE_IS_FILE;
1916                 if (is_zero_pfn(pte_pfn(pte)))
1917                         categories |= PAGE_IS_PFNZERO;
1918                 if (pte_soft_dirty(pte))
1919                         categories |= PAGE_IS_SOFT_DIRTY;
1920         } else if (is_swap_pte(pte)) {
1921                 categories |= PAGE_IS_SWAPPED;
1922                 if (!pte_swp_uffd_wp_any(pte))
1923                         categories |= PAGE_IS_WRITTEN;
1924                 if (pte_swp_soft_dirty(pte))
1925                         categories |= PAGE_IS_SOFT_DIRTY;
1926         }
1927
1928         return categories;
1929 }
1930
1931 static void make_uffd_wp_huge_pte(struct vm_area_struct *vma,
1932                                   unsigned long addr, pte_t *ptep,
1933                                   pte_t ptent)
1934 {
1935         unsigned long psize;
1936
1937         if (is_hugetlb_entry_hwpoisoned(ptent) || is_pte_marker(ptent))
1938                 return;
1939
1940         psize = huge_page_size(hstate_vma(vma));
1941
1942         if (is_hugetlb_entry_migration(ptent))
1943                 set_huge_pte_at(vma->vm_mm, addr, ptep,
1944                                 pte_swp_mkuffd_wp(ptent), psize);
1945         else if (!huge_pte_none(ptent))
1946                 huge_ptep_modify_prot_commit(vma, addr, ptep, ptent,
1947                                              huge_pte_mkuffd_wp(ptent));
1948         else
1949                 set_huge_pte_at(vma->vm_mm, addr, ptep,
1950                                 make_pte_marker(PTE_MARKER_UFFD_WP), psize);
1951 }
1952 #endif /* CONFIG_HUGETLB_PAGE */
1953
1954 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE)
1955 static void pagemap_scan_backout_range(struct pagemap_scan_private *p,
1956                                        unsigned long addr, unsigned long end)
1957 {
1958         struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
1959
1960         if (cur_buf->start != addr)
1961                 cur_buf->end = addr;
1962         else
1963                 cur_buf->start = cur_buf->end = 0;
1964
1965         p->found_pages -= (end - addr) / PAGE_SIZE;
1966 }
1967 #endif
1968
1969 static bool pagemap_scan_is_interesting_page(unsigned long categories,
1970                                              const struct pagemap_scan_private *p)
1971 {
1972         categories ^= p->arg.category_inverted;
1973         if ((categories & p->arg.category_mask) != p->arg.category_mask)
1974                 return false;
1975         if (p->arg.category_anyof_mask && !(categories & p->arg.category_anyof_mask))
1976                 return false;
1977
1978         return true;
1979 }
1980
1981 static bool pagemap_scan_is_interesting_vma(unsigned long categories,
1982                                             const struct pagemap_scan_private *p)
1983 {
1984         unsigned long required = p->arg.category_mask & PAGE_IS_WPALLOWED;
1985
1986         categories ^= p->arg.category_inverted;
1987         if ((categories & required) != required)
1988                 return false;
1989
1990         return true;
1991 }
1992
1993 static int pagemap_scan_test_walk(unsigned long start, unsigned long end,
1994                                   struct mm_walk *walk)
1995 {
1996         struct pagemap_scan_private *p = walk->private;
1997         struct vm_area_struct *vma = walk->vma;
1998         unsigned long vma_category = 0;
1999         bool wp_allowed = userfaultfd_wp_async(vma) &&
2000             userfaultfd_wp_use_markers(vma);
2001
2002         if (!wp_allowed) {
2003                 /* User requested explicit failure over wp-async capability */
2004                 if (p->arg.flags & PM_SCAN_CHECK_WPASYNC)
2005                         return -EPERM;
2006                 /*
2007                  * User requires wr-protect, and allows silently skipping
2008                  * unsupported vmas.
2009                  */
2010                 if (p->arg.flags & PM_SCAN_WP_MATCHING)
2011                         return 1;
2012                 /*
2013                  * Then the request doesn't involve wr-protects at all,
2014                  * fall through to the rest checks, and allow vma walk.
2015                  */
2016         }
2017
2018         if (vma->vm_flags & VM_PFNMAP)
2019                 return 1;
2020
2021         if (wp_allowed)
2022                 vma_category |= PAGE_IS_WPALLOWED;
2023
2024         if (vma->vm_flags & VM_SOFTDIRTY)
2025                 vma_category |= PAGE_IS_SOFT_DIRTY;
2026
2027         if (!pagemap_scan_is_interesting_vma(vma_category, p))
2028                 return 1;
2029
2030         p->cur_vma_category = vma_category;
2031
2032         return 0;
2033 }
2034
2035 static bool pagemap_scan_push_range(unsigned long categories,
2036                                     struct pagemap_scan_private *p,
2037                                     unsigned long addr, unsigned long end)
2038 {
2039         struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
2040
2041         /*
2042          * When there is no output buffer provided at all, the sentinel values
2043          * won't match here. There is no other way for `cur_buf->end` to be
2044          * non-zero other than it being non-empty.
2045          */
2046         if (addr == cur_buf->end && categories == cur_buf->categories) {
2047                 cur_buf->end = end;
2048                 return true;
2049         }
2050
2051         if (cur_buf->end) {
2052                 if (p->vec_buf_index >= p->vec_buf_len - 1)
2053                         return false;
2054
2055                 cur_buf = &p->vec_buf[++p->vec_buf_index];
2056         }
2057
2058         cur_buf->start = addr;
2059         cur_buf->end = end;
2060         cur_buf->categories = categories;
2061
2062         return true;
2063 }
2064
2065 static int pagemap_scan_output(unsigned long categories,
2066                                struct pagemap_scan_private *p,
2067                                unsigned long addr, unsigned long *end)
2068 {
2069         unsigned long n_pages, total_pages;
2070         int ret = 0;
2071
2072         if (!p->vec_buf)
2073                 return 0;
2074
2075         categories &= p->arg.return_mask;
2076
2077         n_pages = (*end - addr) / PAGE_SIZE;
2078         if (check_add_overflow(p->found_pages, n_pages, &total_pages) ||
2079             total_pages > p->arg.max_pages) {
2080                 size_t n_too_much = total_pages - p->arg.max_pages;
2081                 *end -= n_too_much * PAGE_SIZE;
2082                 n_pages -= n_too_much;
2083                 ret = -ENOSPC;
2084         }
2085
2086         if (!pagemap_scan_push_range(categories, p, addr, *end)) {
2087                 *end = addr;
2088                 n_pages = 0;
2089                 ret = -ENOSPC;
2090         }
2091
2092         p->found_pages += n_pages;
2093         if (ret)
2094                 p->arg.walk_end = *end;
2095
2096         return ret;
2097 }
2098
2099 static int pagemap_scan_thp_entry(pmd_t *pmd, unsigned long start,
2100                                   unsigned long end, struct mm_walk *walk)
2101 {
2102 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2103         struct pagemap_scan_private *p = walk->private;
2104         struct vm_area_struct *vma = walk->vma;
2105         unsigned long categories;
2106         spinlock_t *ptl;
2107         int ret = 0;
2108
2109         ptl = pmd_trans_huge_lock(pmd, vma);
2110         if (!ptl)
2111                 return -ENOENT;
2112
2113         categories = p->cur_vma_category |
2114                      pagemap_thp_category(p, vma, start, *pmd);
2115
2116         if (!pagemap_scan_is_interesting_page(categories, p))
2117                 goto out_unlock;
2118
2119         ret = pagemap_scan_output(categories, p, start, &end);
2120         if (start == end)
2121                 goto out_unlock;
2122
2123         if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2124                 goto out_unlock;
2125         if (~categories & PAGE_IS_WRITTEN)
2126                 goto out_unlock;
2127
2128         /*
2129          * Break huge page into small pages if the WP operation
2130          * needs to be performed on a portion of the huge page.
2131          */
2132         if (end != start + HPAGE_SIZE) {
2133                 spin_unlock(ptl);
2134                 split_huge_pmd(vma, pmd, start);
2135                 pagemap_scan_backout_range(p, start, end);
2136                 /* Report as if there was no THP */
2137                 return -ENOENT;
2138         }
2139
2140         make_uffd_wp_pmd(vma, start, pmd);
2141         flush_tlb_range(vma, start, end);
2142 out_unlock:
2143         spin_unlock(ptl);
2144         return ret;
2145 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
2146         return -ENOENT;
2147 #endif
2148 }
2149
2150 static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start,
2151                                   unsigned long end, struct mm_walk *walk)
2152 {
2153         struct pagemap_scan_private *p = walk->private;
2154         struct vm_area_struct *vma = walk->vma;
2155         unsigned long addr, flush_end = 0;
2156         pte_t *pte, *start_pte;
2157         spinlock_t *ptl;
2158         int ret;
2159
2160         arch_enter_lazy_mmu_mode();
2161
2162         ret = pagemap_scan_thp_entry(pmd, start, end, walk);
2163         if (ret != -ENOENT) {
2164                 arch_leave_lazy_mmu_mode();
2165                 return ret;
2166         }
2167
2168         ret = 0;
2169         start_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl);
2170         if (!pte) {
2171                 arch_leave_lazy_mmu_mode();
2172                 walk->action = ACTION_AGAIN;
2173                 return 0;
2174         }
2175
2176         if ((p->arg.flags & PM_SCAN_WP_MATCHING) && !p->vec_out) {
2177                 /* Fast path for performing exclusive WP */
2178                 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2179                         if (pte_uffd_wp(ptep_get(pte)))
2180                                 continue;
2181                         make_uffd_wp_pte(vma, addr, pte);
2182                         if (!flush_end)
2183                                 start = addr;
2184                         flush_end = addr + PAGE_SIZE;
2185                 }
2186                 goto flush_and_return;
2187         }
2188
2189         if (!p->arg.category_anyof_mask && !p->arg.category_inverted &&
2190             p->arg.category_mask == PAGE_IS_WRITTEN &&
2191             p->arg.return_mask == PAGE_IS_WRITTEN) {
2192                 for (addr = start; addr < end; pte++, addr += PAGE_SIZE) {
2193                         unsigned long next = addr + PAGE_SIZE;
2194
2195                         if (pte_uffd_wp(ptep_get(pte)))
2196                                 continue;
2197                         ret = pagemap_scan_output(p->cur_vma_category | PAGE_IS_WRITTEN,
2198                                                   p, addr, &next);
2199                         if (next == addr)
2200                                 break;
2201                         if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2202                                 continue;
2203                         make_uffd_wp_pte(vma, addr, pte);
2204                         if (!flush_end)
2205                                 start = addr;
2206                         flush_end = next;
2207                 }
2208                 goto flush_and_return;
2209         }
2210
2211         for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2212                 unsigned long categories = p->cur_vma_category |
2213                                            pagemap_page_category(p, vma, addr, ptep_get(pte));
2214                 unsigned long next = addr + PAGE_SIZE;
2215
2216                 if (!pagemap_scan_is_interesting_page(categories, p))
2217                         continue;
2218
2219                 ret = pagemap_scan_output(categories, p, addr, &next);
2220                 if (next == addr)
2221                         break;
2222
2223                 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2224                         continue;
2225                 if (~categories & PAGE_IS_WRITTEN)
2226                         continue;
2227
2228                 make_uffd_wp_pte(vma, addr, pte);
2229                 if (!flush_end)
2230                         start = addr;
2231                 flush_end = next;
2232         }
2233
2234 flush_and_return:
2235         if (flush_end)
2236                 flush_tlb_range(vma, start, addr);
2237
2238         pte_unmap_unlock(start_pte, ptl);
2239         arch_leave_lazy_mmu_mode();
2240
2241         cond_resched();
2242         return ret;
2243 }
2244
2245 #ifdef CONFIG_HUGETLB_PAGE
2246 static int pagemap_scan_hugetlb_entry(pte_t *ptep, unsigned long hmask,
2247                                       unsigned long start, unsigned long end,
2248                                       struct mm_walk *walk)
2249 {
2250         struct pagemap_scan_private *p = walk->private;
2251         struct vm_area_struct *vma = walk->vma;
2252         unsigned long categories;
2253         spinlock_t *ptl;
2254         int ret = 0;
2255         pte_t pte;
2256
2257         if (~p->arg.flags & PM_SCAN_WP_MATCHING) {
2258                 /* Go the short route when not write-protecting pages. */
2259
2260                 pte = huge_ptep_get(ptep);
2261                 categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2262
2263                 if (!pagemap_scan_is_interesting_page(categories, p))
2264                         return 0;
2265
2266                 return pagemap_scan_output(categories, p, start, &end);
2267         }
2268
2269         i_mmap_lock_write(vma->vm_file->f_mapping);
2270         ptl = huge_pte_lock(hstate_vma(vma), vma->vm_mm, ptep);
2271
2272         pte = huge_ptep_get(ptep);
2273         categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2274
2275         if (!pagemap_scan_is_interesting_page(categories, p))
2276                 goto out_unlock;
2277
2278         ret = pagemap_scan_output(categories, p, start, &end);
2279         if (start == end)
2280                 goto out_unlock;
2281
2282         if (~categories & PAGE_IS_WRITTEN)
2283                 goto out_unlock;
2284
2285         if (end != start + HPAGE_SIZE) {
2286                 /* Partial HugeTLB page WP isn't possible. */
2287                 pagemap_scan_backout_range(p, start, end);
2288                 p->arg.walk_end = start;
2289                 ret = 0;
2290                 goto out_unlock;
2291         }
2292
2293         make_uffd_wp_huge_pte(vma, start, ptep, pte);
2294         flush_hugetlb_tlb_range(vma, start, end);
2295
2296 out_unlock:
2297         spin_unlock(ptl);
2298         i_mmap_unlock_write(vma->vm_file->f_mapping);
2299
2300         return ret;
2301 }
2302 #else
2303 #define pagemap_scan_hugetlb_entry NULL
2304 #endif
2305
2306 static int pagemap_scan_pte_hole(unsigned long addr, unsigned long end,
2307                                  int depth, struct mm_walk *walk)
2308 {
2309         struct pagemap_scan_private *p = walk->private;
2310         struct vm_area_struct *vma = walk->vma;
2311         int ret, err;
2312
2313         if (!vma || !pagemap_scan_is_interesting_page(p->cur_vma_category, p))
2314                 return 0;
2315
2316         ret = pagemap_scan_output(p->cur_vma_category, p, addr, &end);
2317         if (addr == end)
2318                 return ret;
2319
2320         if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2321                 return ret;
2322
2323         err = uffd_wp_range(vma, addr, end - addr, true);
2324         if (err < 0)
2325                 ret = err;
2326
2327         return ret;
2328 }
2329
2330 static const struct mm_walk_ops pagemap_scan_ops = {
2331         .test_walk = pagemap_scan_test_walk,
2332         .pmd_entry = pagemap_scan_pmd_entry,
2333         .pte_hole = pagemap_scan_pte_hole,
2334         .hugetlb_entry = pagemap_scan_hugetlb_entry,
2335 };
2336
2337 static int pagemap_scan_get_args(struct pm_scan_arg *arg,
2338                                  unsigned long uarg)
2339 {
2340         if (copy_from_user(arg, (void __user *)uarg, sizeof(*arg)))
2341                 return -EFAULT;
2342
2343         if (arg->size != sizeof(struct pm_scan_arg))
2344                 return -EINVAL;
2345
2346         /* Validate requested features */
2347         if (arg->flags & ~PM_SCAN_FLAGS)
2348                 return -EINVAL;
2349         if ((arg->category_inverted | arg->category_mask |
2350              arg->category_anyof_mask | arg->return_mask) & ~PM_SCAN_CATEGORIES)
2351                 return -EINVAL;
2352
2353         arg->start = untagged_addr((unsigned long)arg->start);
2354         arg->end = untagged_addr((unsigned long)arg->end);
2355         arg->vec = untagged_addr((unsigned long)arg->vec);
2356
2357         /* Validate memory pointers */
2358         if (!IS_ALIGNED(arg->start, PAGE_SIZE))
2359                 return -EINVAL;
2360         if (!access_ok((void __user *)(long)arg->start, arg->end - arg->start))
2361                 return -EFAULT;
2362         if (!arg->vec && arg->vec_len)
2363                 return -EINVAL;
2364         if (arg->vec && !access_ok((void __user *)(long)arg->vec,
2365                               arg->vec_len * sizeof(struct page_region)))
2366                 return -EFAULT;
2367
2368         /* Fixup default values */
2369         arg->end = ALIGN(arg->end, PAGE_SIZE);
2370         arg->walk_end = 0;
2371         if (!arg->max_pages)
2372                 arg->max_pages = ULONG_MAX;
2373
2374         return 0;
2375 }
2376
2377 static int pagemap_scan_writeback_args(struct pm_scan_arg *arg,
2378                                        unsigned long uargl)
2379 {
2380         struct pm_scan_arg __user *uarg = (void __user *)uargl;
2381
2382         if (copy_to_user(&uarg->walk_end, &arg->walk_end, sizeof(arg->walk_end)))
2383                 return -EFAULT;
2384
2385         return 0;
2386 }
2387
2388 static int pagemap_scan_init_bounce_buffer(struct pagemap_scan_private *p)
2389 {
2390         if (!p->arg.vec_len)
2391                 return 0;
2392
2393         p->vec_buf_len = min_t(size_t, PAGEMAP_WALK_SIZE >> PAGE_SHIFT,
2394                                p->arg.vec_len);
2395         p->vec_buf = kmalloc_array(p->vec_buf_len, sizeof(*p->vec_buf),
2396                                    GFP_KERNEL);
2397         if (!p->vec_buf)
2398                 return -ENOMEM;
2399
2400         p->vec_buf->start = p->vec_buf->end = 0;
2401         p->vec_out = (struct page_region __user *)(long)p->arg.vec;
2402
2403         return 0;
2404 }
2405
2406 static long pagemap_scan_flush_buffer(struct pagemap_scan_private *p)
2407 {
2408         const struct page_region *buf = p->vec_buf;
2409         long n = p->vec_buf_index;
2410
2411         if (!p->vec_buf)
2412                 return 0;
2413
2414         if (buf[n].end != buf[n].start)
2415                 n++;
2416
2417         if (!n)
2418                 return 0;
2419
2420         if (copy_to_user(p->vec_out, buf, n * sizeof(*buf)))
2421                 return -EFAULT;
2422
2423         p->arg.vec_len -= n;
2424         p->vec_out += n;
2425
2426         p->vec_buf_index = 0;
2427         p->vec_buf_len = min_t(size_t, p->vec_buf_len, p->arg.vec_len);
2428         p->vec_buf->start = p->vec_buf->end = 0;
2429
2430         return n;
2431 }
2432
2433 static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg)
2434 {
2435         struct pagemap_scan_private p = {0};
2436         unsigned long walk_start;
2437         size_t n_ranges_out = 0;
2438         int ret;
2439
2440         ret = pagemap_scan_get_args(&p.arg, uarg);
2441         if (ret)
2442                 return ret;
2443
2444         p.masks_of_interest = p.arg.category_mask | p.arg.category_anyof_mask |
2445                               p.arg.return_mask;
2446         ret = pagemap_scan_init_bounce_buffer(&p);
2447         if (ret)
2448                 return ret;
2449
2450         for (walk_start = p.arg.start; walk_start < p.arg.end;
2451                         walk_start = p.arg.walk_end) {
2452                 struct mmu_notifier_range range;
2453                 long n_out;
2454
2455                 if (fatal_signal_pending(current)) {
2456                         ret = -EINTR;
2457                         break;
2458                 }
2459
2460                 ret = mmap_read_lock_killable(mm);
2461                 if (ret)
2462                         break;
2463
2464                 /* Protection change for the range is going to happen. */
2465                 if (p.arg.flags & PM_SCAN_WP_MATCHING) {
2466                         mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_VMA, 0,
2467                                                 mm, walk_start, p.arg.end);
2468                         mmu_notifier_invalidate_range_start(&range);
2469                 }
2470
2471                 ret = walk_page_range(mm, walk_start, p.arg.end,
2472                                       &pagemap_scan_ops, &p);
2473
2474                 if (p.arg.flags & PM_SCAN_WP_MATCHING)
2475                         mmu_notifier_invalidate_range_end(&range);
2476
2477                 mmap_read_unlock(mm);
2478
2479                 n_out = pagemap_scan_flush_buffer(&p);
2480                 if (n_out < 0)
2481                         ret = n_out;
2482                 else
2483                         n_ranges_out += n_out;
2484
2485                 if (ret != -ENOSPC)
2486                         break;
2487
2488                 if (p.arg.vec_len == 0 || p.found_pages == p.arg.max_pages)
2489                         break;
2490         }
2491
2492         /* ENOSPC signifies early stop (buffer full) from the walk. */
2493         if (!ret || ret == -ENOSPC)
2494                 ret = n_ranges_out;
2495
2496         /* The walk_end isn't set when ret is zero */
2497         if (!p.arg.walk_end)
2498                 p.arg.walk_end = p.arg.end;
2499         if (pagemap_scan_writeback_args(&p.arg, uarg))
2500                 ret = -EFAULT;
2501
2502         kfree(p.vec_buf);
2503         return ret;
2504 }
2505
2506 static long do_pagemap_cmd(struct file *file, unsigned int cmd,
2507                            unsigned long arg)
2508 {
2509         struct mm_struct *mm = file->private_data;
2510
2511         switch (cmd) {
2512         case PAGEMAP_SCAN:
2513                 return do_pagemap_scan(mm, arg);
2514
2515         default:
2516                 return -EINVAL;
2517         }
2518 }
2519
2520 const struct file_operations proc_pagemap_operations = {
2521         .llseek         = mem_lseek, /* borrow this */
2522         .read           = pagemap_read,
2523         .open           = pagemap_open,
2524         .release        = pagemap_release,
2525         .unlocked_ioctl = do_pagemap_cmd,
2526         .compat_ioctl   = do_pagemap_cmd,
2527 };
2528 #endif /* CONFIG_PROC_PAGE_MONITOR */
2529
2530 #ifdef CONFIG_NUMA
2531
2532 struct numa_maps {
2533         unsigned long pages;
2534         unsigned long anon;
2535         unsigned long active;
2536         unsigned long writeback;
2537         unsigned long mapcount_max;
2538         unsigned long dirty;
2539         unsigned long swapcache;
2540         unsigned long node[MAX_NUMNODES];
2541 };
2542
2543 struct numa_maps_private {
2544         struct proc_maps_private proc_maps;
2545         struct numa_maps md;
2546 };
2547
2548 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
2549                         unsigned long nr_pages)
2550 {
2551         int count = page_mapcount(page);
2552
2553         md->pages += nr_pages;
2554         if (pte_dirty || PageDirty(page))
2555                 md->dirty += nr_pages;
2556
2557         if (PageSwapCache(page))
2558                 md->swapcache += nr_pages;
2559
2560         if (PageActive(page) || PageUnevictable(page))
2561                 md->active += nr_pages;
2562
2563         if (PageWriteback(page))
2564                 md->writeback += nr_pages;
2565
2566         if (PageAnon(page))
2567                 md->anon += nr_pages;
2568
2569         if (count > md->mapcount_max)
2570                 md->mapcount_max = count;
2571
2572         md->node[page_to_nid(page)] += nr_pages;
2573 }
2574
2575 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
2576                 unsigned long addr)
2577 {
2578         struct page *page;
2579         int nid;
2580
2581         if (!pte_present(pte))
2582                 return NULL;
2583
2584         page = vm_normal_page(vma, addr, pte);
2585         if (!page || is_zone_device_page(page))
2586                 return NULL;
2587
2588         if (PageReserved(page))
2589                 return NULL;
2590
2591         nid = page_to_nid(page);
2592         if (!node_isset(nid, node_states[N_MEMORY]))
2593                 return NULL;
2594
2595         return page;
2596 }
2597
2598 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2599 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
2600                                               struct vm_area_struct *vma,
2601                                               unsigned long addr)
2602 {
2603         struct page *page;
2604         int nid;
2605
2606         if (!pmd_present(pmd))
2607                 return NULL;
2608
2609         page = vm_normal_page_pmd(vma, addr, pmd);
2610         if (!page)
2611                 return NULL;
2612
2613         if (PageReserved(page))
2614                 return NULL;
2615
2616         nid = page_to_nid(page);
2617         if (!node_isset(nid, node_states[N_MEMORY]))
2618                 return NULL;
2619
2620         return page;
2621 }
2622 #endif
2623
2624 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
2625                 unsigned long end, struct mm_walk *walk)
2626 {
2627         struct numa_maps *md = walk->private;
2628         struct vm_area_struct *vma = walk->vma;
2629         spinlock_t *ptl;
2630         pte_t *orig_pte;
2631         pte_t *pte;
2632
2633 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2634         ptl = pmd_trans_huge_lock(pmd, vma);
2635         if (ptl) {
2636                 struct page *page;
2637
2638                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
2639                 if (page)
2640                         gather_stats(page, md, pmd_dirty(*pmd),
2641                                      HPAGE_PMD_SIZE/PAGE_SIZE);
2642                 spin_unlock(ptl);
2643                 return 0;
2644         }
2645 #endif
2646         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
2647         if (!pte) {
2648                 walk->action = ACTION_AGAIN;
2649                 return 0;
2650         }
2651         do {
2652                 pte_t ptent = ptep_get(pte);
2653                 struct page *page = can_gather_numa_stats(ptent, vma, addr);
2654                 if (!page)
2655                         continue;
2656                 gather_stats(page, md, pte_dirty(ptent), 1);
2657
2658         } while (pte++, addr += PAGE_SIZE, addr != end);
2659         pte_unmap_unlock(orig_pte, ptl);
2660         cond_resched();
2661         return 0;
2662 }
2663 #ifdef CONFIG_HUGETLB_PAGE
2664 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
2665                 unsigned long addr, unsigned long end, struct mm_walk *walk)
2666 {
2667         pte_t huge_pte = huge_ptep_get(pte);
2668         struct numa_maps *md;
2669         struct page *page;
2670
2671         if (!pte_present(huge_pte))
2672                 return 0;
2673
2674         page = pte_page(huge_pte);
2675
2676         md = walk->private;
2677         gather_stats(page, md, pte_dirty(huge_pte), 1);
2678         return 0;
2679 }
2680
2681 #else
2682 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
2683                 unsigned long addr, unsigned long end, struct mm_walk *walk)
2684 {
2685         return 0;
2686 }
2687 #endif
2688
2689 static const struct mm_walk_ops show_numa_ops = {
2690         .hugetlb_entry = gather_hugetlb_stats,
2691         .pmd_entry = gather_pte_stats,
2692         .walk_lock = PGWALK_RDLOCK,
2693 };
2694
2695 /*
2696  * Display pages allocated per node and memory policy via /proc.
2697  */
2698 static int show_numa_map(struct seq_file *m, void *v)
2699 {
2700         struct numa_maps_private *numa_priv = m->private;
2701         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
2702         struct vm_area_struct *vma = v;
2703         struct numa_maps *md = &numa_priv->md;
2704         struct file *file = vma->vm_file;
2705         struct mm_struct *mm = vma->vm_mm;
2706         char buffer[64];
2707         struct mempolicy *pol;
2708         pgoff_t ilx;
2709         int nid;
2710
2711         if (!mm)
2712                 return 0;
2713
2714         /* Ensure we start with an empty set of numa_maps statistics. */
2715         memset(md, 0, sizeof(*md));
2716
2717         pol = __get_vma_policy(vma, vma->vm_start, &ilx);
2718         if (pol) {
2719                 mpol_to_str(buffer, sizeof(buffer), pol);
2720                 mpol_cond_put(pol);
2721         } else {
2722                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
2723         }
2724
2725         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
2726
2727         if (file) {
2728                 seq_puts(m, " file=");
2729                 seq_path(m, file_user_path(file), "\n\t= ");
2730         } else if (vma_is_initial_heap(vma)) {
2731                 seq_puts(m, " heap");
2732         } else if (vma_is_initial_stack(vma)) {
2733                 seq_puts(m, " stack");
2734         }
2735
2736         if (is_vm_hugetlb_page(vma))
2737                 seq_puts(m, " huge");
2738
2739         /* mmap_lock is held by m_start */
2740         walk_page_vma(vma, &show_numa_ops, md);
2741
2742         if (!md->pages)
2743                 goto out;
2744
2745         if (md->anon)
2746                 seq_printf(m, " anon=%lu", md->anon);
2747
2748         if (md->dirty)
2749                 seq_printf(m, " dirty=%lu", md->dirty);
2750
2751         if (md->pages != md->anon && md->pages != md->dirty)
2752                 seq_printf(m, " mapped=%lu", md->pages);
2753
2754         if (md->mapcount_max > 1)
2755                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
2756
2757         if (md->swapcache)
2758                 seq_printf(m, " swapcache=%lu", md->swapcache);
2759
2760         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2761                 seq_printf(m, " active=%lu", md->active);
2762
2763         if (md->writeback)
2764                 seq_printf(m, " writeback=%lu", md->writeback);
2765
2766         for_each_node_state(nid, N_MEMORY)
2767                 if (md->node[nid])
2768                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2769
2770         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2771 out:
2772         seq_putc(m, '\n');
2773         return 0;
2774 }
2775
2776 static const struct seq_operations proc_pid_numa_maps_op = {
2777         .start  = m_start,
2778         .next   = m_next,
2779         .stop   = m_stop,
2780         .show   = show_numa_map,
2781 };
2782
2783 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2784 {
2785         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2786                                 sizeof(struct numa_maps_private));
2787 }
2788
2789 const struct file_operations proc_pid_numa_maps_operations = {
2790         .open           = pid_numa_maps_open,
2791         .read           = seq_read,
2792         .llseek         = seq_lseek,
2793         .release        = proc_map_release,
2794 };
2795
2796 #endif /* CONFIG_NUMA */