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 folio *folio, unsigned long size, unsigned long pss,
 415                 bool dirty, bool locked, bool private)
 416 {
 417         mss->pss += pss;
 418
 419         if (folio_test_anon(folio))
 420                 mss->pss_anon += pss;
 421         else if (folio_test_swapbacked(folio))
 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 || folio_test_dirty(folio)) {
 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         struct folio *folio = page_folio(page);
 448         int i, nr = compound ? compound_nr(page) : 1;
 449         unsigned long size = nr * PAGE_SIZE;
 450
 451         /*
 452          * First accumulate quantities that depend only on |size| and the type
 453          * of the compound page.
 454          */
 455         if (folio_test_anon(folio)) {
 456                 mss->anonymous += size;
 457                 if (!folio_test_swapbacked(folio) && !dirty &&
 458                     !folio_test_dirty(folio))
 459                         mss->lazyfree += size;
 460         }
 461
 462         if (folio_test_ksm(folio))
 463                 mss->ksm += size;
 464
 465         mss->resident += size;
 466         /* Accumulate the size in pages that have been accessed. */
 467         if (young || folio_test_young(folio) || folio_test_referenced(folio))
 468                 mss->referenced += size;
 469
 470         /*
 471          * Then accumulate quantities that may depend on sharing, or that may
 472          * differ page-by-page.
 473          *
 474          * refcount == 1 guarantees the page is mapped exactly once.
 475          * If any subpage of the compound page mapped with PTE it would elevate
 476          * the refcount.
 477          *
 478          * The page_mapcount() is called to get a snapshot of the mapcount.
 479          * Without holding the page lock this snapshot can be slightly wrong as
 480          * we cannot always read the mapcount atomically.  It is not safe to
 481          * call page_mapcount() even with PTL held if the page is not mapped,
 482          * especially for migration entries.  Treat regular migration entries
 483          * as mapcount == 1.
 484          */
 485         if ((folio_ref_count(folio) == 1) || migration) {
 486                 smaps_page_accumulate(mss, folio, size, size << PSS_SHIFT,
 487                                 dirty, locked, true);
 488                 return;
 489         }
 490         for (i = 0; i < nr; i++, page++) {
 491                 int mapcount = page_mapcount(page);
 492                 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
 493                 if (mapcount >= 2)
 494                         pss /= mapcount;
 495                 smaps_page_accumulate(mss, folio, PAGE_SIZE, pss,
 496                                 dirty, locked, mapcount < 2);
 497         }
 498 }
 499
 500 #ifdef CONFIG_SHMEM
 501 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 502                           __always_unused int depth, struct mm_walk *walk)
 503 {
 504         struct mem_size_stats *mss = walk->private;
 505         struct vm_area_struct *vma = walk->vma;
 506
 507         mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping,
 508                                               linear_page_index(vma, addr),
 509                                               linear_page_index(vma, end));
 510
 511         return 0;
 512 }
 513 #else
 514 #define smaps_pte_hole          NULL
 515 #endif /* CONFIG_SHMEM */
 516
 517 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
 518 {
 519 #ifdef CONFIG_SHMEM
 520         if (walk->ops->pte_hole) {
 521                 /* depth is not used */
 522                 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
 523         }
 524 #endif
 525 }
 526
 527 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 528                 struct mm_walk *walk)
 529 {
 530         struct mem_size_stats *mss = walk->private;
 531         struct vm_area_struct *vma = walk->vma;
 532         bool locked = !!(vma->vm_flags & VM_LOCKED);
 533         struct page *page = NULL;
 534         bool migration = false, young = false, dirty = false;
 535         pte_t ptent = ptep_get(pte);
 536
 537         if (pte_present(ptent)) {
 538                 page = vm_normal_page(vma, addr, ptent);
 539                 young = pte_young(ptent);
 540                 dirty = pte_dirty(ptent);
 541         } else if (is_swap_pte(ptent)) {
 542                 swp_entry_t swpent = pte_to_swp_entry(ptent);
 543
 544                 if (!non_swap_entry(swpent)) {
 545                         int mapcount;
 546
 547                         mss->swap += PAGE_SIZE;
 548                         mapcount = swp_swapcount(swpent);
 549                         if (mapcount >= 2) {
 550                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 551
 552                                 do_div(pss_delta, mapcount);
 553                                 mss->swap_pss += pss_delta;
 554                         } else {
 555                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 556                         }
 557                 } else if (is_pfn_swap_entry(swpent)) {
 558                         if (is_migration_entry(swpent))
 559                                 migration = true;
 560                         page = pfn_swap_entry_to_page(swpent);
 561                 }
 562         } else {
 563                 smaps_pte_hole_lookup(addr, walk);
 564                 return;
 565         }
 566
 567         if (!page)
 568                 return;
 569
 570         smaps_account(mss, page, false, young, dirty, locked, migration);
 571 }
 572
 573 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 574 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 575                 struct mm_walk *walk)
 576 {
 577         struct mem_size_stats *mss = walk->private;
 578         struct vm_area_struct *vma = walk->vma;
 579         bool locked = !!(vma->vm_flags & VM_LOCKED);
 580         struct page *page = NULL;
 581         struct folio *folio;
 582         bool migration = false;
 583
 584         if (pmd_present(*pmd)) {
 585                 page = vm_normal_page_pmd(vma, addr, *pmd);
 586         } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
 587                 swp_entry_t entry = pmd_to_swp_entry(*pmd);
 588
 589                 if (is_migration_entry(entry)) {
 590                         migration = true;
 591                         page = pfn_swap_entry_to_page(entry);
 592                 }
 593         }
 594         if (IS_ERR_OR_NULL(page))
 595                 return;
 596         folio = page_folio(page);
 597         if (folio_test_anon(folio))
 598                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 599         else if (folio_test_swapbacked(folio))
 600                 mss->shmem_thp += HPAGE_PMD_SIZE;
 601         else if (folio_is_zone_device(folio))
 602                 /* pass */;
 603         else
 604                 mss->file_thp += HPAGE_PMD_SIZE;
 605
 606         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
 607                       locked, migration);
 608 }
 609 #else
 610 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 611                 struct mm_walk *walk)
 612 {
 613 }
 614 #endif
 615
 616 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 617                            struct mm_walk *walk)
 618 {
 619         struct vm_area_struct *vma = walk->vma;
 620         pte_t *pte;
 621         spinlock_t *ptl;
 622
 623         ptl = pmd_trans_huge_lock(pmd, vma);
 624         if (ptl) {
 625                 smaps_pmd_entry(pmd, addr, walk);
 626                 spin_unlock(ptl);
 627                 goto out;
 628         }
 629
 630         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 631         if (!pte) {
 632                 walk->action = ACTION_AGAIN;
 633                 return 0;
 634         }
 635         for (; addr != end; pte++, addr += PAGE_SIZE)
 636                 smaps_pte_entry(pte, addr, walk);
 637         pte_unmap_unlock(pte - 1, ptl);
 638 out:
 639         cond_resched();
 640         return 0;
 641 }
 642
 643 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 644 {
 645         /*
 646          * Don't forget to update Documentation/ on changes.
 647          */
 648         static const char mnemonics[BITS_PER_LONG][2] = {
 649                 /*
 650                  * In case if we meet a flag we don't know about.
 651                  */
 652                 [0 ... (BITS_PER_LONG-1)] = "??",
 653
 654                 [ilog2(VM_READ)]        = "rd",
 655                 [ilog2(VM_WRITE)]       = "wr",
 656                 [ilog2(VM_EXEC)]        = "ex",
 657                 [ilog2(VM_SHARED)]      = "sh",
 658                 [ilog2(VM_MAYREAD)]     = "mr",
 659                 [ilog2(VM_MAYWRITE)]    = "mw",
 660                 [ilog2(VM_MAYEXEC)]     = "me",
 661                 [ilog2(VM_MAYSHARE)]    = "ms",
 662                 [ilog2(VM_GROWSDOWN)]   = "gd",
 663                 [ilog2(VM_PFNMAP)]      = "pf",
 664                 [ilog2(VM_LOCKED)]      = "lo",
 665                 [ilog2(VM_IO)]          = "io",
 666                 [ilog2(VM_SEQ_READ)]    = "sr",
 667                 [ilog2(VM_RAND_READ)]   = "rr",
 668                 [ilog2(VM_DONTCOPY)]    = "dc",
 669                 [ilog2(VM_DONTEXPAND)]  = "de",
 670                 [ilog2(VM_LOCKONFAULT)] = "lf",
 671                 [ilog2(VM_ACCOUNT)]     = "ac",
 672                 [ilog2(VM_NORESERVE)]   = "nr",
 673                 [ilog2(VM_HUGETLB)]     = "ht",
 674                 [ilog2(VM_SYNC)]        = "sf",
 675                 [ilog2(VM_ARCH_1)]      = "ar",
 676                 [ilog2(VM_WIPEONFORK)]  = "wf",
 677                 [ilog2(VM_DONTDUMP)]    = "dd",
 678 #ifdef CONFIG_ARM64_BTI
 679                 [ilog2(VM_ARM64_BTI)]   = "bt",
 680 #endif
 681 #ifdef CONFIG_MEM_SOFT_DIRTY
 682                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 683 #endif
 684                 [ilog2(VM_MIXEDMAP)]    = "mm",
 685                 [ilog2(VM_HUGEPAGE)]    = "hg",
 686                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 687                 [ilog2(VM_MERGEABLE)]   = "mg",
 688                 [ilog2(VM_UFFD_MISSING)]= "um",
 689                 [ilog2(VM_UFFD_WP)]     = "uw",
 690 #ifdef CONFIG_ARM64_MTE
 691                 [ilog2(VM_MTE)]         = "mt",
 692                 [ilog2(VM_MTE_ALLOWED)] = "",
 693 #endif
 694 #ifdef CONFIG_ARCH_HAS_PKEYS
 695                 /* These come out via ProtectionKey: */
 696                 [ilog2(VM_PKEY_BIT0)]   = "",
 697                 [ilog2(VM_PKEY_BIT1)]   = "",
 698                 [ilog2(VM_PKEY_BIT2)]   = "",
 699                 [ilog2(VM_PKEY_BIT3)]   = "",
 700 #if VM_PKEY_BIT4
 701                 [ilog2(VM_PKEY_BIT4)]   = "",
 702 #endif
 703 #endif /* CONFIG_ARCH_HAS_PKEYS */
 704 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
 705                 [ilog2(VM_UFFD_MINOR)]  = "ui",
 706 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
 707 #ifdef CONFIG_X86_USER_SHADOW_STACK
 708                 [ilog2(VM_SHADOW_STACK)] = "ss",
 709 #endif
 710 #ifdef CONFIG_64BIT
 711                 [ilog2(VM_SEALED)] = "sl",
 712 #endif
 713         };
 714         size_t i;
 715
 716         seq_puts(m, "VmFlags: ");
 717         for (i = 0; i < BITS_PER_LONG; i++) {
 718                 if (!mnemonics[i][0])
 719                         continue;
 720                 if (vma->vm_flags & (1UL << i)) {
 721                         seq_putc(m, mnemonics[i][0]);
 722                         seq_putc(m, mnemonics[i][1]);
 723                         seq_putc(m, ' ');
 724                 }
 725         }
 726         seq_putc(m, '\n');
 727 }
 728
 729 #ifdef CONFIG_HUGETLB_PAGE
 730 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 731                                  unsigned long addr, unsigned long end,
 732                                  struct mm_walk *walk)
 733 {
 734         struct mem_size_stats *mss = walk->private;
 735         struct vm_area_struct *vma = walk->vma;
 736         pte_t ptent = huge_ptep_get(pte);
 737         struct folio *folio = NULL;
 738
 739         if (pte_present(ptent)) {
 740                 folio = page_folio(pte_page(ptent));
 741         } else if (is_swap_pte(ptent)) {
 742                 swp_entry_t swpent = pte_to_swp_entry(ptent);
 743
 744                 if (is_pfn_swap_entry(swpent))
 745                         folio = pfn_swap_entry_folio(swpent);
 746         }
 747         if (folio) {
 748                 if (folio_likely_mapped_shared(folio) ||
 749                     hugetlb_pmd_shared(pte))
 750                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 751                 else
 752                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 753         }
 754         return 0;
 755 }
 756 #else
 757 #define smaps_hugetlb_range     NULL
 758 #endif /* HUGETLB_PAGE */
 759
 760 static const struct mm_walk_ops smaps_walk_ops = {
 761         .pmd_entry              = smaps_pte_range,
 762         .hugetlb_entry          = smaps_hugetlb_range,
 763         .walk_lock              = PGWALK_RDLOCK,
 764 };
 765
 766 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 767         .pmd_entry              = smaps_pte_range,
 768         .hugetlb_entry          = smaps_hugetlb_range,
 769         .pte_hole               = smaps_pte_hole,
 770         .walk_lock              = PGWALK_RDLOCK,
 771 };
 772
 773 /*
 774  * Gather mem stats from @vma with the indicated beginning
 775  * address @start, and keep them in @mss.
 776  *
 777  * Use vm_start of @vma as the beginning address if @start is 0.
 778  */
 779 static void smap_gather_stats(struct vm_area_struct *vma,
 780                 struct mem_size_stats *mss, unsigned long start)
 781 {
 782         const struct mm_walk_ops *ops = &smaps_walk_ops;
 783
 784         /* Invalid start */
 785         if (start >= vma->vm_end)
 786                 return;
 787
 788         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 789                 /*
 790                  * For shared or readonly shmem mappings we know that all
 791                  * swapped out pages belong to the shmem object, and we can
 792                  * obtain the swap value much more efficiently. For private
 793                  * writable mappings, we might have COW pages that are
 794                  * not affected by the parent swapped out pages of the shmem
 795                  * object, so we have to distinguish them during the page walk.
 796                  * Unless we know that the shmem object (or the part mapped by
 797                  * our VMA) has no swapped out pages at all.
 798                  */
 799                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 800
 801                 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 802                                         !(vma->vm_flags & VM_WRITE))) {
 803                         mss->swap += shmem_swapped;
 804                 } else {
 805                         ops = &smaps_shmem_walk_ops;
 806                 }
 807         }
 808
 809         /* mmap_lock is held in m_start */
 810         if (!start)
 811                 walk_page_vma(vma, ops, mss);
 812         else
 813                 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
 814 }
 815
 816 #define SEQ_PUT_DEC(str, val) \
 817                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 818
 819 /* Show the contents common for smaps and smaps_rollup */
 820 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 821         bool rollup_mode)
 822 {
 823         SEQ_PUT_DEC("Rss:            ", mss->resident);
 824         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 825         SEQ_PUT_DEC(" kB\nPss_Dirty:      ", mss->pss_dirty >> PSS_SHIFT);
 826         if (rollup_mode) {
 827                 /*
 828                  * These are meaningful only for smaps_rollup, otherwise two of
 829                  * them are zero, and the other one is the same as Pss.
 830                  */
 831                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 832                         mss->pss_anon >> PSS_SHIFT);
 833                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 834                         mss->pss_file >> PSS_SHIFT);
 835                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 836                         mss->pss_shmem >> PSS_SHIFT);
 837         }
 838         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 839         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 840         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 841         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 842         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 843         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 844         SEQ_PUT_DEC(" kB\nKSM:            ", mss->ksm);
 845         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 846         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 847         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 848         SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
 849         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 850         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 851                                   mss->private_hugetlb >> 10, 7);
 852         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 853         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 854                                         mss->swap_pss >> PSS_SHIFT);
 855         SEQ_PUT_DEC(" kB\nLocked:         ",
 856                                         mss->pss_locked >> PSS_SHIFT);
 857         seq_puts(m, " kB\n");
 858 }
 859
 860 static int show_smap(struct seq_file *m, void *v)
 861 {
 862         struct vm_area_struct *vma = v;
 863         struct mem_size_stats mss = {};
 864
 865         smap_gather_stats(vma, &mss, 0);
 866
 867         show_map_vma(m, vma);
 868
 869         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 870         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 871         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 872         seq_puts(m, " kB\n");
 873
 874         __show_smap(m, &mss, false);
 875
 876         seq_printf(m, "THPeligible:    %8u\n",
 877                    !!thp_vma_allowable_orders(vma, vma->vm_flags,
 878                            TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
 879
 880         if (arch_pkeys_enabled())
 881                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 882         show_smap_vma_flags(m, vma);
 883
 884         return 0;
 885 }
 886
 887 static int show_smaps_rollup(struct seq_file *m, void *v)
 888 {
 889         struct proc_maps_private *priv = m->private;
 890         struct mem_size_stats mss = {};
 891         struct mm_struct *mm = priv->mm;
 892         struct vm_area_struct *vma;
 893         unsigned long vma_start = 0, last_vma_end = 0;
 894         int ret = 0;
 895         VMA_ITERATOR(vmi, mm, 0);
 896
 897         priv->task = get_proc_task(priv->inode);
 898         if (!priv->task)
 899                 return -ESRCH;
 900
 901         if (!mm || !mmget_not_zero(mm)) {
 902                 ret = -ESRCH;
 903                 goto out_put_task;
 904         }
 905
 906         ret = mmap_read_lock_killable(mm);
 907         if (ret)
 908                 goto out_put_mm;
 909
 910         hold_task_mempolicy(priv);
 911         vma = vma_next(&vmi);
 912
 913         if (unlikely(!vma))
 914                 goto empty_set;
 915
 916         vma_start = vma->vm_start;
 917         do {
 918                 smap_gather_stats(vma, &mss, 0);
 919                 last_vma_end = vma->vm_end;
 920
 921                 /*
 922                  * Release mmap_lock temporarily if someone wants to
 923                  * access it for write request.
 924                  */
 925                 if (mmap_lock_is_contended(mm)) {
 926                         vma_iter_invalidate(&vmi);
 927                         mmap_read_unlock(mm);
 928                         ret = mmap_read_lock_killable(mm);
 929                         if (ret) {
 930                                 release_task_mempolicy(priv);
 931                                 goto out_put_mm;
 932                         }
 933
 934                         /*
 935                          * After dropping the lock, there are four cases to
 936                          * consider. See the following example for explanation.
 937                          *
 938                          *   +------+------+-----------+
 939                          *   | VMA1 | VMA2 | VMA3      |
 940                          *   +------+------+-----------+
 941                          *   |      |      |           |
 942                          *  4k     8k     16k         400k
 943                          *
 944                          * Suppose we drop the lock after reading VMA2 due to
 945                          * contention, then we get:
 946                          *
 947                          *      last_vma_end = 16k
 948                          *
 949                          * 1) VMA2 is freed, but VMA3 exists:
 950                          *
 951                          *    vma_next(vmi) will return VMA3.
 952                          *    In this case, just continue from VMA3.
 953                          *
 954                          * 2) VMA2 still exists:
 955                          *
 956                          *    vma_next(vmi) will return VMA3.
 957                          *    In this case, just continue from VMA3.
 958                          *
 959                          * 3) No more VMAs can be found:
 960                          *
 961                          *    vma_next(vmi) will return NULL.
 962                          *    No more things to do, just break.
 963                          *
 964                          * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
 965                          *
 966                          *    vma_next(vmi) will return VMA' whose range
 967                          *    contains last_vma_end.
 968                          *    Iterate VMA' from last_vma_end.
 969                          */
 970                         vma = vma_next(&vmi);
 971                         /* Case 3 above */
 972                         if (!vma)
 973                                 break;
 974
 975                         /* Case 1 and 2 above */
 976                         if (vma->vm_start >= last_vma_end) {
 977                                 smap_gather_stats(vma, &mss, 0);
 978                                 last_vma_end = vma->vm_end;
 979                                 continue;
 980                         }
 981
 982                         /* Case 4 above */
 983                         if (vma->vm_end > last_vma_end) {
 984                                 smap_gather_stats(vma, &mss, last_vma_end);
 985                                 last_vma_end = vma->vm_end;
 986                         }
 987                 }
 988         } for_each_vma(vmi, vma);
 989
 990 empty_set:
 991         show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
 992         seq_pad(m, ' ');
 993         seq_puts(m, "[rollup]\n");
 994
 995         __show_smap(m, &mss, true);
 996
 997         release_task_mempolicy(priv);
 998         mmap_read_unlock(mm);
 999
1000 out_put_mm:
1001         mmput(mm);
1002 out_put_task:
1003         put_task_struct(priv->task);
1004         priv->task = NULL;
1005
1006         return ret;
1007 }
1008 #undef SEQ_PUT_DEC
1009
1010 static const struct seq_operations proc_pid_smaps_op = {
1011         .start  = m_start,
1012         .next   = m_next,
1013         .stop   = m_stop,
1014         .show   = show_smap
1015 };
1016
1017 static int pid_smaps_open(struct inode *inode, struct file *file)
1018 {
1019         return do_maps_open(inode, file, &proc_pid_smaps_op);
1020 }
1021
1022 static int smaps_rollup_open(struct inode *inode, struct file *file)
1023 {
1024         int ret;
1025         struct proc_maps_private *priv;
1026
1027         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1028         if (!priv)
1029                 return -ENOMEM;
1030
1031         ret = single_open(file, show_smaps_rollup, priv);
1032         if (ret)
1033                 goto out_free;
1034
1035         priv->inode = inode;
1036         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1037         if (IS_ERR(priv->mm)) {
1038                 ret = PTR_ERR(priv->mm);
1039
1040                 single_release(inode, file);
1041                 goto out_free;
1042         }
1043
1044         return 0;
1045
1046 out_free:
1047         kfree(priv);
1048         return ret;
1049 }
1050
1051 static int smaps_rollup_release(struct inode *inode, struct file *file)
1052 {
1053         struct seq_file *seq = file->private_data;
1054         struct proc_maps_private *priv = seq->private;
1055
1056         if (priv->mm)
1057                 mmdrop(priv->mm);
1058
1059         kfree(priv);
1060         return single_release(inode, file);
1061 }
1062
1063 const struct file_operations proc_pid_smaps_operations = {
1064         .open           = pid_smaps_open,
1065         .read           = seq_read,
1066         .llseek         = seq_lseek,
1067         .release        = proc_map_release,
1068 };
1069
1070 const struct file_operations proc_pid_smaps_rollup_operations = {
1071         .open           = smaps_rollup_open,
1072         .read           = seq_read,
1073         .llseek         = seq_lseek,
1074         .release        = smaps_rollup_release,
1075 };
1076
1077 enum clear_refs_types {
1078         CLEAR_REFS_ALL = 1,
1079         CLEAR_REFS_ANON,
1080         CLEAR_REFS_MAPPED,
1081         CLEAR_REFS_SOFT_DIRTY,
1082         CLEAR_REFS_MM_HIWATER_RSS,
1083         CLEAR_REFS_LAST,
1084 };
1085
1086 struct clear_refs_private {
1087         enum clear_refs_types type;
1088 };
1089
1090 #ifdef CONFIG_MEM_SOFT_DIRTY
1091
1092 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1093 {
1094         struct page *page;
1095
1096         if (!pte_write(pte))
1097                 return false;
1098         if (!is_cow_mapping(vma->vm_flags))
1099                 return false;
1100         if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1101                 return false;
1102         page = vm_normal_page(vma, addr, pte);
1103         if (!page)
1104                 return false;
1105         return page_maybe_dma_pinned(page);
1106 }
1107
1108 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1109                 unsigned long addr, pte_t *pte)
1110 {
1111         /*
1112          * The soft-dirty tracker uses #PF-s to catch writes
1113          * to pages, so write-protect the pte as well. See the
1114          * Documentation/admin-guide/mm/soft-dirty.rst for full description
1115          * of how soft-dirty works.
1116          */
1117         pte_t ptent = ptep_get(pte);
1118
1119         if (pte_present(ptent)) {
1120                 pte_t old_pte;
1121
1122                 if (pte_is_pinned(vma, addr, ptent))
1123                         return;
1124                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1125                 ptent = pte_wrprotect(old_pte);
1126                 ptent = pte_clear_soft_dirty(ptent);
1127                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1128         } else if (is_swap_pte(ptent)) {
1129                 ptent = pte_swp_clear_soft_dirty(ptent);
1130                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1131         }
1132 }
1133 #else
1134 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1135                 unsigned long addr, pte_t *pte)
1136 {
1137 }
1138 #endif
1139
1140 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1141 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1142                 unsigned long addr, pmd_t *pmdp)
1143 {
1144         pmd_t old, pmd = *pmdp;
1145
1146         if (pmd_present(pmd)) {
1147                 /* See comment in change_huge_pmd() */
1148                 old = pmdp_invalidate(vma, addr, pmdp);
1149                 if (pmd_dirty(old))
1150                         pmd = pmd_mkdirty(pmd);
1151                 if (pmd_young(old))
1152                         pmd = pmd_mkyoung(pmd);
1153
1154                 pmd = pmd_wrprotect(pmd);
1155                 pmd = pmd_clear_soft_dirty(pmd);
1156
1157                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1158         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1159                 pmd = pmd_swp_clear_soft_dirty(pmd);
1160                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1161         }
1162 }
1163 #else
1164 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1165                 unsigned long addr, pmd_t *pmdp)
1166 {
1167 }
1168 #endif
1169
1170 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1171                                 unsigned long end, struct mm_walk *walk)
1172 {
1173         struct clear_refs_private *cp = walk->private;
1174         struct vm_area_struct *vma = walk->vma;
1175         pte_t *pte, ptent;
1176         spinlock_t *ptl;
1177         struct folio *folio;
1178
1179         ptl = pmd_trans_huge_lock(pmd, vma);
1180         if (ptl) {
1181                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1182                         clear_soft_dirty_pmd(vma, addr, pmd);
1183                         goto out;
1184                 }
1185
1186                 if (!pmd_present(*pmd))
1187                         goto out;
1188
1189                 folio = pmd_folio(*pmd);
1190
1191                 /* Clear accessed and referenced bits. */
1192                 pmdp_test_and_clear_young(vma, addr, pmd);
1193                 folio_test_clear_young(folio);
1194                 folio_clear_referenced(folio);
1195 out:
1196                 spin_unlock(ptl);
1197                 return 0;
1198         }
1199
1200         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1201         if (!pte) {
1202                 walk->action = ACTION_AGAIN;
1203                 return 0;
1204         }
1205         for (; addr != end; pte++, addr += PAGE_SIZE) {
1206                 ptent = ptep_get(pte);
1207
1208                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1209                         clear_soft_dirty(vma, addr, pte);
1210                         continue;
1211                 }
1212
1213                 if (!pte_present(ptent))
1214                         continue;
1215
1216                 folio = vm_normal_folio(vma, addr, ptent);
1217                 if (!folio)
1218                         continue;
1219
1220                 /* Clear accessed and referenced bits. */
1221                 ptep_test_and_clear_young(vma, addr, pte);
1222                 folio_test_clear_young(folio);
1223                 folio_clear_referenced(folio);
1224         }
1225         pte_unmap_unlock(pte - 1, ptl);
1226         cond_resched();
1227         return 0;
1228 }
1229
1230 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1231                                 struct mm_walk *walk)
1232 {
1233         struct clear_refs_private *cp = walk->private;
1234         struct vm_area_struct *vma = walk->vma;
1235
1236         if (vma->vm_flags & VM_PFNMAP)
1237                 return 1;
1238
1239         /*
1240          * Writing 1 to /proc/pid/clear_refs affects all pages.
1241          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1242          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1243          * Writing 4 to /proc/pid/clear_refs affects all pages.
1244          */
1245         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1246                 return 1;
1247         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1248                 return 1;
1249         return 0;
1250 }
1251
1252 static const struct mm_walk_ops clear_refs_walk_ops = {
1253         .pmd_entry              = clear_refs_pte_range,
1254         .test_walk              = clear_refs_test_walk,
1255         .walk_lock              = PGWALK_WRLOCK,
1256 };
1257
1258 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1259                                 size_t count, loff_t *ppos)
1260 {
1261         struct task_struct *task;
1262         char buffer[PROC_NUMBUF] = {};
1263         struct mm_struct *mm;
1264         struct vm_area_struct *vma;
1265         enum clear_refs_types type;
1266         int itype;
1267         int rv;
1268
1269         if (count > sizeof(buffer) - 1)
1270                 count = sizeof(buffer) - 1;
1271         if (copy_from_user(buffer, buf, count))
1272                 return -EFAULT;
1273         rv = kstrtoint(strstrip(buffer), 10, &itype);
1274         if (rv < 0)
1275                 return rv;
1276         type = (enum clear_refs_types)itype;
1277         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1278                 return -EINVAL;
1279
1280         task = get_proc_task(file_inode(file));
1281         if (!task)
1282                 return -ESRCH;
1283         mm = get_task_mm(task);
1284         if (mm) {
1285                 VMA_ITERATOR(vmi, mm, 0);
1286                 struct mmu_notifier_range range;
1287                 struct clear_refs_private cp = {
1288                         .type = type,
1289                 };
1290
1291                 if (mmap_write_lock_killable(mm)) {
1292                         count = -EINTR;
1293                         goto out_mm;
1294                 }
1295                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1296                         /*
1297                          * Writing 5 to /proc/pid/clear_refs resets the peak
1298                          * resident set size to this mm's current rss value.
1299                          */
1300                         reset_mm_hiwater_rss(mm);
1301                         goto out_unlock;
1302                 }
1303
1304                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1305                         for_each_vma(vmi, vma) {
1306                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1307                                         continue;
1308                                 vm_flags_clear(vma, VM_SOFTDIRTY);
1309                                 vma_set_page_prot(vma);
1310                         }
1311
1312                         inc_tlb_flush_pending(mm);
1313                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1314                                                 0, mm, 0, -1UL);
1315                         mmu_notifier_invalidate_range_start(&range);
1316                 }
1317                 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1318                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1319                         mmu_notifier_invalidate_range_end(&range);
1320                         flush_tlb_mm(mm);
1321                         dec_tlb_flush_pending(mm);
1322                 }
1323 out_unlock:
1324                 mmap_write_unlock(mm);
1325 out_mm:
1326                 mmput(mm);
1327         }
1328         put_task_struct(task);
1329
1330         return count;
1331 }
1332
1333 const struct file_operations proc_clear_refs_operations = {
1334         .write          = clear_refs_write,
1335         .llseek         = noop_llseek,
1336 };
1337
1338 typedef struct {
1339         u64 pme;
1340 } pagemap_entry_t;
1341
1342 struct pagemapread {
1343         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1344         pagemap_entry_t *buffer;
1345         bool show_pfn;
1346 };
1347
1348 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1349 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1350
1351 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1352 #define PM_PFRAME_BITS          55
1353 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1354 #define PM_SOFT_DIRTY           BIT_ULL(55)
1355 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1356 #define PM_UFFD_WP              BIT_ULL(57)
1357 #define PM_FILE                 BIT_ULL(61)
1358 #define PM_SWAP                 BIT_ULL(62)
1359 #define PM_PRESENT              BIT_ULL(63)
1360
1361 #define PM_END_OF_BUFFER    1
1362
1363 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1364 {
1365         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1366 }
1367
1368 static int add_to_pagemap(pagemap_entry_t *pme, struct pagemapread *pm)
1369 {
1370         pm->buffer[pm->pos++] = *pme;
1371         if (pm->pos >= pm->len)
1372                 return PM_END_OF_BUFFER;
1373         return 0;
1374 }
1375
1376 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1377                             __always_unused int depth, struct mm_walk *walk)
1378 {
1379         struct pagemapread *pm = walk->private;
1380         unsigned long addr = start;
1381         int err = 0;
1382
1383         while (addr < end) {
1384                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1385                 pagemap_entry_t pme = make_pme(0, 0);
1386                 /* End of address space hole, which we mark as non-present. */
1387                 unsigned long hole_end;
1388
1389                 if (vma)
1390                         hole_end = min(end, vma->vm_start);
1391                 else
1392                         hole_end = end;
1393
1394                 for (; addr < hole_end; addr += PAGE_SIZE) {
1395                         err = add_to_pagemap(&pme, pm);
1396                         if (err)
1397                                 goto out;
1398                 }
1399
1400                 if (!vma)
1401                         break;
1402
1403                 /* Addresses in the VMA. */
1404                 if (vma->vm_flags & VM_SOFTDIRTY)
1405                         pme = make_pme(0, PM_SOFT_DIRTY);
1406                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1407                         err = add_to_pagemap(&pme, pm);
1408                         if (err)
1409                                 goto out;
1410                 }
1411         }
1412 out:
1413         return err;
1414 }
1415
1416 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1417                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1418 {
1419         u64 frame = 0, flags = 0;
1420         struct page *page = NULL;
1421         bool migration = false;
1422
1423         if (pte_present(pte)) {
1424                 if (pm->show_pfn)
1425                         frame = pte_pfn(pte);
1426                 flags |= PM_PRESENT;
1427                 page = vm_normal_page(vma, addr, pte);
1428                 if (pte_soft_dirty(pte))
1429                         flags |= PM_SOFT_DIRTY;
1430                 if (pte_uffd_wp(pte))
1431                         flags |= PM_UFFD_WP;
1432         } else if (is_swap_pte(pte)) {
1433                 swp_entry_t entry;
1434                 if (pte_swp_soft_dirty(pte))
1435                         flags |= PM_SOFT_DIRTY;
1436                 if (pte_swp_uffd_wp(pte))
1437                         flags |= PM_UFFD_WP;
1438                 entry = pte_to_swp_entry(pte);
1439                 if (pm->show_pfn) {
1440                         pgoff_t offset;
1441                         /*
1442                          * For PFN swap offsets, keeping the offset field
1443                          * to be PFN only to be compatible with old smaps.
1444                          */
1445                         if (is_pfn_swap_entry(entry))
1446                                 offset = swp_offset_pfn(entry);
1447                         else
1448                                 offset = swp_offset(entry);
1449                         frame = swp_type(entry) |
1450                             (offset << MAX_SWAPFILES_SHIFT);
1451                 }
1452                 flags |= PM_SWAP;
1453                 migration = is_migration_entry(entry);
1454                 if (is_pfn_swap_entry(entry))
1455                         page = pfn_swap_entry_to_page(entry);
1456                 if (pte_marker_entry_uffd_wp(entry))
1457                         flags |= PM_UFFD_WP;
1458         }
1459
1460         if (page && !PageAnon(page))
1461                 flags |= PM_FILE;
1462         if (page && !migration && page_mapcount(page) == 1)
1463                 flags |= PM_MMAP_EXCLUSIVE;
1464         if (vma->vm_flags & VM_SOFTDIRTY)
1465                 flags |= PM_SOFT_DIRTY;
1466
1467         return make_pme(frame, flags);
1468 }
1469
1470 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1471                              struct mm_walk *walk)
1472 {
1473         struct vm_area_struct *vma = walk->vma;
1474         struct pagemapread *pm = walk->private;
1475         spinlock_t *ptl;
1476         pte_t *pte, *orig_pte;
1477         int err = 0;
1478 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1479         bool migration = false;
1480
1481         ptl = pmd_trans_huge_lock(pmdp, vma);
1482         if (ptl) {
1483                 u64 flags = 0, frame = 0;
1484                 pmd_t pmd = *pmdp;
1485                 struct page *page = NULL;
1486
1487                 if (vma->vm_flags & VM_SOFTDIRTY)
1488                         flags |= PM_SOFT_DIRTY;
1489
1490                 if (pmd_present(pmd)) {
1491                         page = pmd_page(pmd);
1492
1493                         flags |= PM_PRESENT;
1494                         if (pmd_soft_dirty(pmd))
1495                                 flags |= PM_SOFT_DIRTY;
1496                         if (pmd_uffd_wp(pmd))
1497                                 flags |= PM_UFFD_WP;
1498                         if (pm->show_pfn)
1499                                 frame = pmd_pfn(pmd) +
1500                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1501                 }
1502 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1503                 else if (is_swap_pmd(pmd)) {
1504                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1505                         unsigned long offset;
1506
1507                         if (pm->show_pfn) {
1508                                 if (is_pfn_swap_entry(entry))
1509                                         offset = swp_offset_pfn(entry);
1510                                 else
1511                                         offset = swp_offset(entry);
1512                                 offset = offset +
1513                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1514                                 frame = swp_type(entry) |
1515                                         (offset << MAX_SWAPFILES_SHIFT);
1516                         }
1517                         flags |= PM_SWAP;
1518                         if (pmd_swp_soft_dirty(pmd))
1519                                 flags |= PM_SOFT_DIRTY;
1520                         if (pmd_swp_uffd_wp(pmd))
1521                                 flags |= PM_UFFD_WP;
1522                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1523                         migration = is_migration_entry(entry);
1524                         page = pfn_swap_entry_to_page(entry);
1525                 }
1526 #endif
1527
1528                 if (page && !migration && page_mapcount(page) == 1)
1529                         flags |= PM_MMAP_EXCLUSIVE;
1530
1531                 for (; addr != end; addr += PAGE_SIZE) {
1532                         pagemap_entry_t pme = make_pme(frame, flags);
1533
1534                         err = add_to_pagemap(&pme, pm);
1535                         if (err)
1536                                 break;
1537                         if (pm->show_pfn) {
1538                                 if (flags & PM_PRESENT)
1539                                         frame++;
1540                                 else if (flags & PM_SWAP)
1541                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1542                         }
1543                 }
1544                 spin_unlock(ptl);
1545                 return err;
1546         }
1547 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1548
1549         /*
1550          * We can assume that @vma always points to a valid one and @end never
1551          * goes beyond vma->vm_end.
1552          */
1553         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1554         if (!pte) {
1555                 walk->action = ACTION_AGAIN;
1556                 return err;
1557         }
1558         for (; addr < end; pte++, addr += PAGE_SIZE) {
1559                 pagemap_entry_t pme;
1560
1561                 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte));
1562                 err = add_to_pagemap(&pme, pm);
1563                 if (err)
1564                         break;
1565         }
1566         pte_unmap_unlock(orig_pte, ptl);
1567
1568         cond_resched();
1569
1570         return err;
1571 }
1572
1573 #ifdef CONFIG_HUGETLB_PAGE
1574 /* This function walks within one hugetlb entry in the single call */
1575 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1576                                  unsigned long addr, unsigned long end,
1577                                  struct mm_walk *walk)
1578 {
1579         struct pagemapread *pm = walk->private;
1580         struct vm_area_struct *vma = walk->vma;
1581         u64 flags = 0, frame = 0;
1582         int err = 0;
1583         pte_t pte;
1584
1585         if (vma->vm_flags & VM_SOFTDIRTY)
1586                 flags |= PM_SOFT_DIRTY;
1587
1588         pte = huge_ptep_get(ptep);
1589         if (pte_present(pte)) {
1590                 struct folio *folio = page_folio(pte_page(pte));
1591
1592                 if (!folio_test_anon(folio))
1593                         flags |= PM_FILE;
1594
1595                 if (!folio_likely_mapped_shared(folio) &&
1596                     !hugetlb_pmd_shared(ptep))
1597                         flags |= PM_MMAP_EXCLUSIVE;
1598
1599                 if (huge_pte_uffd_wp(pte))
1600                         flags |= PM_UFFD_WP;
1601
1602                 flags |= PM_PRESENT;
1603                 if (pm->show_pfn)
1604                         frame = pte_pfn(pte) +
1605                                 ((addr & ~hmask) >> PAGE_SHIFT);
1606         } else if (pte_swp_uffd_wp_any(pte)) {
1607                 flags |= PM_UFFD_WP;
1608         }
1609
1610         for (; addr != end; addr += PAGE_SIZE) {
1611                 pagemap_entry_t pme = make_pme(frame, flags);
1612
1613                 err = add_to_pagemap(&pme, pm);
1614                 if (err)
1615                         return err;
1616                 if (pm->show_pfn && (flags & PM_PRESENT))
1617                         frame++;
1618         }
1619
1620         cond_resched();
1621
1622         return err;
1623 }
1624 #else
1625 #define pagemap_hugetlb_range   NULL
1626 #endif /* HUGETLB_PAGE */
1627
1628 static const struct mm_walk_ops pagemap_ops = {
1629         .pmd_entry      = pagemap_pmd_range,
1630         .pte_hole       = pagemap_pte_hole,
1631         .hugetlb_entry  = pagemap_hugetlb_range,
1632         .walk_lock      = PGWALK_RDLOCK,
1633 };
1634
1635 /*
1636  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1637  *
1638  * For each page in the address space, this file contains one 64-bit entry
1639  * consisting of the following:
1640  *
1641  * Bits 0-54  page frame number (PFN) if present
1642  * Bits 0-4   swap type if swapped
1643  * Bits 5-54  swap offset if swapped
1644  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1645  * Bit  56    page exclusively mapped
1646  * Bit  57    pte is uffd-wp write-protected
1647  * Bits 58-60 zero
1648  * Bit  61    page is file-page or shared-anon
1649  * Bit  62    page swapped
1650  * Bit  63    page present
1651  *
1652  * If the page is not present but in swap, then the PFN contains an
1653  * encoding of the swap file number and the page's offset into the
1654  * swap. Unmapped pages return a null PFN. This allows determining
1655  * precisely which pages are mapped (or in swap) and comparing mapped
1656  * pages between processes.
1657  *
1658  * Efficient users of this interface will use /proc/pid/maps to
1659  * determine which areas of memory are actually mapped and llseek to
1660  * skip over unmapped regions.
1661  */
1662 static ssize_t pagemap_read(struct file *file, char __user *buf,
1663                             size_t count, loff_t *ppos)
1664 {
1665         struct mm_struct *mm = file->private_data;
1666         struct pagemapread pm;
1667         unsigned long src;
1668         unsigned long svpfn;
1669         unsigned long start_vaddr;
1670         unsigned long end_vaddr;
1671         int ret = 0, copied = 0;
1672
1673         if (!mm || !mmget_not_zero(mm))
1674                 goto out;
1675
1676         ret = -EINVAL;
1677         /* file position must be aligned */
1678         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1679                 goto out_mm;
1680
1681         ret = 0;
1682         if (!count)
1683                 goto out_mm;
1684
1685         /* do not disclose physical addresses: attack vector */
1686         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1687
1688         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1689         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1690         ret = -ENOMEM;
1691         if (!pm.buffer)
1692                 goto out_mm;
1693
1694         src = *ppos;
1695         svpfn = src / PM_ENTRY_BYTES;
1696         end_vaddr = mm->task_size;
1697
1698         /* watch out for wraparound */
1699         start_vaddr = end_vaddr;
1700         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
1701                 unsigned long end;
1702
1703                 ret = mmap_read_lock_killable(mm);
1704                 if (ret)
1705                         goto out_free;
1706                 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
1707                 mmap_read_unlock(mm);
1708
1709                 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
1710                 if (end >= start_vaddr && end < mm->task_size)
1711                         end_vaddr = end;
1712         }
1713
1714         /* Ensure the address is inside the task */
1715         if (start_vaddr > mm->task_size)
1716                 start_vaddr = end_vaddr;
1717
1718         ret = 0;
1719         while (count && (start_vaddr < end_vaddr)) {
1720                 int len;
1721                 unsigned long end;
1722
1723                 pm.pos = 0;
1724                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1725                 /* overflow ? */
1726                 if (end < start_vaddr || end > end_vaddr)
1727                         end = end_vaddr;
1728                 ret = mmap_read_lock_killable(mm);
1729                 if (ret)
1730                         goto out_free;
1731                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1732                 mmap_read_unlock(mm);
1733                 start_vaddr = end;
1734
1735                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1736                 if (copy_to_user(buf, pm.buffer, len)) {
1737                         ret = -EFAULT;
1738                         goto out_free;
1739                 }
1740                 copied += len;
1741                 buf += len;
1742                 count -= len;
1743         }
1744         *ppos += copied;
1745         if (!ret || ret == PM_END_OF_BUFFER)
1746                 ret = copied;
1747
1748 out_free:
1749         kfree(pm.buffer);
1750 out_mm:
1751         mmput(mm);
1752 out:
1753         return ret;
1754 }
1755
1756 static int pagemap_open(struct inode *inode, struct file *file)
1757 {
1758         struct mm_struct *mm;
1759
1760         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1761         if (IS_ERR(mm))
1762                 return PTR_ERR(mm);
1763         file->private_data = mm;
1764         return 0;
1765 }
1766
1767 static int pagemap_release(struct inode *inode, struct file *file)
1768 {
1769         struct mm_struct *mm = file->private_data;
1770
1771         if (mm)
1772                 mmdrop(mm);
1773         return 0;
1774 }
1775
1776 #define PM_SCAN_CATEGORIES      (PAGE_IS_WPALLOWED | PAGE_IS_WRITTEN |  \
1777                                  PAGE_IS_FILE | PAGE_IS_PRESENT |       \
1778                                  PAGE_IS_SWAPPED | PAGE_IS_PFNZERO |    \
1779                                  PAGE_IS_HUGE | PAGE_IS_SOFT_DIRTY)
1780 #define PM_SCAN_FLAGS           (PM_SCAN_WP_MATCHING | PM_SCAN_CHECK_WPASYNC)
1781
1782 struct pagemap_scan_private {
1783         struct pm_scan_arg arg;
1784         unsigned long masks_of_interest, cur_vma_category;
1785         struct page_region *vec_buf;
1786         unsigned long vec_buf_len, vec_buf_index, found_pages;
1787         struct page_region __user *vec_out;
1788 };
1789
1790 static unsigned long pagemap_page_category(struct pagemap_scan_private *p,
1791                                            struct vm_area_struct *vma,
1792                                            unsigned long addr, pte_t pte)
1793 {
1794         unsigned long categories = 0;
1795
1796         if (pte_present(pte)) {
1797                 struct page *page;
1798
1799                 categories |= PAGE_IS_PRESENT;
1800                 if (!pte_uffd_wp(pte))
1801                         categories |= PAGE_IS_WRITTEN;
1802
1803                 if (p->masks_of_interest & PAGE_IS_FILE) {
1804                         page = vm_normal_page(vma, addr, pte);
1805                         if (page && !PageAnon(page))
1806                                 categories |= PAGE_IS_FILE;
1807                 }
1808
1809                 if (is_zero_pfn(pte_pfn(pte)))
1810                         categories |= PAGE_IS_PFNZERO;
1811                 if (pte_soft_dirty(pte))
1812                         categories |= PAGE_IS_SOFT_DIRTY;
1813         } else if (is_swap_pte(pte)) {
1814                 swp_entry_t swp;
1815
1816                 categories |= PAGE_IS_SWAPPED;
1817                 if (!pte_swp_uffd_wp_any(pte))
1818                         categories |= PAGE_IS_WRITTEN;
1819
1820                 if (p->masks_of_interest & PAGE_IS_FILE) {
1821                         swp = pte_to_swp_entry(pte);
1822                         if (is_pfn_swap_entry(swp) &&
1823                             !folio_test_anon(pfn_swap_entry_folio(swp)))
1824                                 categories |= PAGE_IS_FILE;
1825                 }
1826                 if (pte_swp_soft_dirty(pte))
1827                         categories |= PAGE_IS_SOFT_DIRTY;
1828         }
1829
1830         return categories;
1831 }
1832
1833 static void make_uffd_wp_pte(struct vm_area_struct *vma,
1834                              unsigned long addr, pte_t *pte, pte_t ptent)
1835 {
1836         if (pte_present(ptent)) {
1837                 pte_t old_pte;
1838
1839                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1840                 ptent = pte_mkuffd_wp(old_pte);
1841                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1842         } else if (is_swap_pte(ptent)) {
1843                 ptent = pte_swp_mkuffd_wp(ptent);
1844                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1845         } else {
1846                 set_pte_at(vma->vm_mm, addr, pte,
1847                            make_pte_marker(PTE_MARKER_UFFD_WP));
1848         }
1849 }
1850
1851 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1852 static unsigned long pagemap_thp_category(struct pagemap_scan_private *p,
1853                                           struct vm_area_struct *vma,
1854                                           unsigned long addr, pmd_t pmd)
1855 {
1856         unsigned long categories = PAGE_IS_HUGE;
1857
1858         if (pmd_present(pmd)) {
1859                 struct page *page;
1860
1861                 categories |= PAGE_IS_PRESENT;
1862                 if (!pmd_uffd_wp(pmd))
1863                         categories |= PAGE_IS_WRITTEN;
1864
1865                 if (p->masks_of_interest & PAGE_IS_FILE) {
1866                         page = vm_normal_page_pmd(vma, addr, pmd);
1867                         if (page && !PageAnon(page))
1868                                 categories |= PAGE_IS_FILE;
1869                 }
1870
1871                 if (is_zero_pfn(pmd_pfn(pmd)))
1872                         categories |= PAGE_IS_PFNZERO;
1873                 if (pmd_soft_dirty(pmd))
1874                         categories |= PAGE_IS_SOFT_DIRTY;
1875         } else if (is_swap_pmd(pmd)) {
1876                 swp_entry_t swp;
1877
1878                 categories |= PAGE_IS_SWAPPED;
1879                 if (!pmd_swp_uffd_wp(pmd))
1880                         categories |= PAGE_IS_WRITTEN;
1881                 if (pmd_swp_soft_dirty(pmd))
1882                         categories |= PAGE_IS_SOFT_DIRTY;
1883
1884                 if (p->masks_of_interest & PAGE_IS_FILE) {
1885                         swp = pmd_to_swp_entry(pmd);
1886                         if (is_pfn_swap_entry(swp) &&
1887                             !folio_test_anon(pfn_swap_entry_folio(swp)))
1888                                 categories |= PAGE_IS_FILE;
1889                 }
1890         }
1891
1892         return categories;
1893 }
1894
1895 static void make_uffd_wp_pmd(struct vm_area_struct *vma,
1896                              unsigned long addr, pmd_t *pmdp)
1897 {
1898         pmd_t old, pmd = *pmdp;
1899
1900         if (pmd_present(pmd)) {
1901                 old = pmdp_invalidate_ad(vma, addr, pmdp);
1902                 pmd = pmd_mkuffd_wp(old);
1903                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1904         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1905                 pmd = pmd_swp_mkuffd_wp(pmd);
1906                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1907         }
1908 }
1909 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1910
1911 #ifdef CONFIG_HUGETLB_PAGE
1912 static unsigned long pagemap_hugetlb_category(pte_t pte)
1913 {
1914         unsigned long categories = PAGE_IS_HUGE;
1915
1916         /*
1917          * According to pagemap_hugetlb_range(), file-backed HugeTLB
1918          * page cannot be swapped. So PAGE_IS_FILE is not checked for
1919          * swapped pages.
1920          */
1921         if (pte_present(pte)) {
1922                 categories |= PAGE_IS_PRESENT;
1923                 if (!huge_pte_uffd_wp(pte))
1924                         categories |= PAGE_IS_WRITTEN;
1925                 if (!PageAnon(pte_page(pte)))
1926                         categories |= PAGE_IS_FILE;
1927                 if (is_zero_pfn(pte_pfn(pte)))
1928                         categories |= PAGE_IS_PFNZERO;
1929                 if (pte_soft_dirty(pte))
1930                         categories |= PAGE_IS_SOFT_DIRTY;
1931         } else if (is_swap_pte(pte)) {
1932                 categories |= PAGE_IS_SWAPPED;
1933                 if (!pte_swp_uffd_wp_any(pte))
1934                         categories |= PAGE_IS_WRITTEN;
1935                 if (pte_swp_soft_dirty(pte))
1936                         categories |= PAGE_IS_SOFT_DIRTY;
1937         }
1938
1939         return categories;
1940 }
1941
1942 static void make_uffd_wp_huge_pte(struct vm_area_struct *vma,
1943                                   unsigned long addr, pte_t *ptep,
1944                                   pte_t ptent)
1945 {
1946         unsigned long psize;
1947
1948         if (is_hugetlb_entry_hwpoisoned(ptent) || is_pte_marker(ptent))
1949                 return;
1950
1951         psize = huge_page_size(hstate_vma(vma));
1952
1953         if (is_hugetlb_entry_migration(ptent))
1954                 set_huge_pte_at(vma->vm_mm, addr, ptep,
1955                                 pte_swp_mkuffd_wp(ptent), psize);
1956         else if (!huge_pte_none(ptent))
1957                 huge_ptep_modify_prot_commit(vma, addr, ptep, ptent,
1958                                              huge_pte_mkuffd_wp(ptent));
1959         else
1960                 set_huge_pte_at(vma->vm_mm, addr, ptep,
1961                                 make_pte_marker(PTE_MARKER_UFFD_WP), psize);
1962 }
1963 #endif /* CONFIG_HUGETLB_PAGE */
1964
1965 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE)
1966 static void pagemap_scan_backout_range(struct pagemap_scan_private *p,
1967                                        unsigned long addr, unsigned long end)
1968 {
1969         struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
1970
1971         if (cur_buf->start != addr)
1972                 cur_buf->end = addr;
1973         else
1974                 cur_buf->start = cur_buf->end = 0;
1975
1976         p->found_pages -= (end - addr) / PAGE_SIZE;
1977 }
1978 #endif
1979
1980 static bool pagemap_scan_is_interesting_page(unsigned long categories,
1981                                              const struct pagemap_scan_private *p)
1982 {
1983         categories ^= p->arg.category_inverted;
1984         if ((categories & p->arg.category_mask) != p->arg.category_mask)
1985                 return false;
1986         if (p->arg.category_anyof_mask && !(categories & p->arg.category_anyof_mask))
1987                 return false;
1988
1989         return true;
1990 }
1991
1992 static bool pagemap_scan_is_interesting_vma(unsigned long categories,
1993                                             const struct pagemap_scan_private *p)
1994 {
1995         unsigned long required = p->arg.category_mask & PAGE_IS_WPALLOWED;
1996
1997         categories ^= p->arg.category_inverted;
1998         if ((categories & required) != required)
1999                 return false;
2000
2001         return true;
2002 }
2003
2004 static int pagemap_scan_test_walk(unsigned long start, unsigned long end,
2005                                   struct mm_walk *walk)
2006 {
2007         struct pagemap_scan_private *p = walk->private;
2008         struct vm_area_struct *vma = walk->vma;
2009         unsigned long vma_category = 0;
2010         bool wp_allowed = userfaultfd_wp_async(vma) &&
2011             userfaultfd_wp_use_markers(vma);
2012
2013         if (!wp_allowed) {
2014                 /* User requested explicit failure over wp-async capability */
2015                 if (p->arg.flags & PM_SCAN_CHECK_WPASYNC)
2016                         return -EPERM;
2017                 /*
2018                  * User requires wr-protect, and allows silently skipping
2019                  * unsupported vmas.
2020                  */
2021                 if (p->arg.flags & PM_SCAN_WP_MATCHING)
2022                         return 1;
2023                 /*
2024                  * Then the request doesn't involve wr-protects at all,
2025                  * fall through to the rest checks, and allow vma walk.
2026                  */
2027         }
2028
2029         if (vma->vm_flags & VM_PFNMAP)
2030                 return 1;
2031
2032         if (wp_allowed)
2033                 vma_category |= PAGE_IS_WPALLOWED;
2034
2035         if (vma->vm_flags & VM_SOFTDIRTY)
2036                 vma_category |= PAGE_IS_SOFT_DIRTY;
2037
2038         if (!pagemap_scan_is_interesting_vma(vma_category, p))
2039                 return 1;
2040
2041         p->cur_vma_category = vma_category;
2042
2043         return 0;
2044 }
2045
2046 static bool pagemap_scan_push_range(unsigned long categories,
2047                                     struct pagemap_scan_private *p,
2048                                     unsigned long addr, unsigned long end)
2049 {
2050         struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
2051
2052         /*
2053          * When there is no output buffer provided at all, the sentinel values
2054          * won't match here. There is no other way for `cur_buf->end` to be
2055          * non-zero other than it being non-empty.
2056          */
2057         if (addr == cur_buf->end && categories == cur_buf->categories) {
2058                 cur_buf->end = end;
2059                 return true;
2060         }
2061
2062         if (cur_buf->end) {
2063                 if (p->vec_buf_index >= p->vec_buf_len - 1)
2064                         return false;
2065
2066                 cur_buf = &p->vec_buf[++p->vec_buf_index];
2067         }
2068
2069         cur_buf->start = addr;
2070         cur_buf->end = end;
2071         cur_buf->categories = categories;
2072
2073         return true;
2074 }
2075
2076 static int pagemap_scan_output(unsigned long categories,
2077                                struct pagemap_scan_private *p,
2078                                unsigned long addr, unsigned long *end)
2079 {
2080         unsigned long n_pages, total_pages;
2081         int ret = 0;
2082
2083         if (!p->vec_buf)
2084                 return 0;
2085
2086         categories &= p->arg.return_mask;
2087
2088         n_pages = (*end - addr) / PAGE_SIZE;
2089         if (check_add_overflow(p->found_pages, n_pages, &total_pages) ||
2090             total_pages > p->arg.max_pages) {
2091                 size_t n_too_much = total_pages - p->arg.max_pages;
2092                 *end -= n_too_much * PAGE_SIZE;
2093                 n_pages -= n_too_much;
2094                 ret = -ENOSPC;
2095         }
2096
2097         if (!pagemap_scan_push_range(categories, p, addr, *end)) {
2098                 *end = addr;
2099                 n_pages = 0;
2100                 ret = -ENOSPC;
2101         }
2102
2103         p->found_pages += n_pages;
2104         if (ret)
2105                 p->arg.walk_end = *end;
2106
2107         return ret;
2108 }
2109
2110 static int pagemap_scan_thp_entry(pmd_t *pmd, unsigned long start,
2111                                   unsigned long end, struct mm_walk *walk)
2112 {
2113 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2114         struct pagemap_scan_private *p = walk->private;
2115         struct vm_area_struct *vma = walk->vma;
2116         unsigned long categories;
2117         spinlock_t *ptl;
2118         int ret = 0;
2119
2120         ptl = pmd_trans_huge_lock(pmd, vma);
2121         if (!ptl)
2122                 return -ENOENT;
2123
2124         categories = p->cur_vma_category |
2125                      pagemap_thp_category(p, vma, start, *pmd);
2126
2127         if (!pagemap_scan_is_interesting_page(categories, p))
2128                 goto out_unlock;
2129
2130         ret = pagemap_scan_output(categories, p, start, &end);
2131         if (start == end)
2132                 goto out_unlock;
2133
2134         if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2135                 goto out_unlock;
2136         if (~categories & PAGE_IS_WRITTEN)
2137                 goto out_unlock;
2138
2139         /*
2140          * Break huge page into small pages if the WP operation
2141          * needs to be performed on a portion of the huge page.
2142          */
2143         if (end != start + HPAGE_SIZE) {
2144                 spin_unlock(ptl);
2145                 split_huge_pmd(vma, pmd, start);
2146                 pagemap_scan_backout_range(p, start, end);
2147                 /* Report as if there was no THP */
2148                 return -ENOENT;
2149         }
2150
2151         make_uffd_wp_pmd(vma, start, pmd);
2152         flush_tlb_range(vma, start, end);
2153 out_unlock:
2154         spin_unlock(ptl);
2155         return ret;
2156 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
2157         return -ENOENT;
2158 #endif
2159 }
2160
2161 static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start,
2162                                   unsigned long end, struct mm_walk *walk)
2163 {
2164         struct pagemap_scan_private *p = walk->private;
2165         struct vm_area_struct *vma = walk->vma;
2166         unsigned long addr, flush_end = 0;
2167         pte_t *pte, *start_pte;
2168         spinlock_t *ptl;
2169         int ret;
2170
2171         arch_enter_lazy_mmu_mode();
2172
2173         ret = pagemap_scan_thp_entry(pmd, start, end, walk);
2174         if (ret != -ENOENT) {
2175                 arch_leave_lazy_mmu_mode();
2176                 return ret;
2177         }
2178
2179         ret = 0;
2180         start_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl);
2181         if (!pte) {
2182                 arch_leave_lazy_mmu_mode();
2183                 walk->action = ACTION_AGAIN;
2184                 return 0;
2185         }
2186
2187         if ((p->arg.flags & PM_SCAN_WP_MATCHING) && !p->vec_out) {
2188                 /* Fast path for performing exclusive WP */
2189                 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2190                         pte_t ptent = ptep_get(pte);
2191
2192                         if ((pte_present(ptent) && pte_uffd_wp(ptent)) ||
2193                             pte_swp_uffd_wp_any(ptent))
2194                                 continue;
2195                         make_uffd_wp_pte(vma, addr, pte, ptent);
2196                         if (!flush_end)
2197                                 start = addr;
2198                         flush_end = addr + PAGE_SIZE;
2199                 }
2200                 goto flush_and_return;
2201         }
2202
2203         if (!p->arg.category_anyof_mask && !p->arg.category_inverted &&
2204             p->arg.category_mask == PAGE_IS_WRITTEN &&
2205             p->arg.return_mask == PAGE_IS_WRITTEN) {
2206                 for (addr = start; addr < end; pte++, addr += PAGE_SIZE) {
2207                         unsigned long next = addr + PAGE_SIZE;
2208                         pte_t ptent = ptep_get(pte);
2209
2210                         if ((pte_present(ptent) && pte_uffd_wp(ptent)) ||
2211                             pte_swp_uffd_wp_any(ptent))
2212                                 continue;
2213                         ret = pagemap_scan_output(p->cur_vma_category | PAGE_IS_WRITTEN,
2214                                                   p, addr, &next);
2215                         if (next == addr)
2216                                 break;
2217                         if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2218                                 continue;
2219                         make_uffd_wp_pte(vma, addr, pte, ptent);
2220                         if (!flush_end)
2221                                 start = addr;
2222                         flush_end = next;
2223                 }
2224                 goto flush_and_return;
2225         }
2226
2227         for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2228                 pte_t ptent = ptep_get(pte);
2229                 unsigned long categories = p->cur_vma_category |
2230                                            pagemap_page_category(p, vma, addr, ptent);
2231                 unsigned long next = addr + PAGE_SIZE;
2232
2233                 if (!pagemap_scan_is_interesting_page(categories, p))
2234                         continue;
2235
2236                 ret = pagemap_scan_output(categories, p, addr, &next);
2237                 if (next == addr)
2238                         break;
2239
2240                 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2241                         continue;
2242                 if (~categories & PAGE_IS_WRITTEN)
2243                         continue;
2244
2245                 make_uffd_wp_pte(vma, addr, pte, ptent);
2246                 if (!flush_end)
2247                         start = addr;
2248                 flush_end = next;
2249         }
2250
2251 flush_and_return:
2252         if (flush_end)
2253                 flush_tlb_range(vma, start, addr);
2254
2255         pte_unmap_unlock(start_pte, ptl);
2256         arch_leave_lazy_mmu_mode();
2257
2258         cond_resched();
2259         return ret;
2260 }
2261
2262 #ifdef CONFIG_HUGETLB_PAGE
2263 static int pagemap_scan_hugetlb_entry(pte_t *ptep, unsigned long hmask,
2264                                       unsigned long start, unsigned long end,
2265                                       struct mm_walk *walk)
2266 {
2267         struct pagemap_scan_private *p = walk->private;
2268         struct vm_area_struct *vma = walk->vma;
2269         unsigned long categories;
2270         spinlock_t *ptl;
2271         int ret = 0;
2272         pte_t pte;
2273
2274         if (~p->arg.flags & PM_SCAN_WP_MATCHING) {
2275                 /* Go the short route when not write-protecting pages. */
2276
2277                 pte = huge_ptep_get(ptep);
2278                 categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2279
2280                 if (!pagemap_scan_is_interesting_page(categories, p))
2281                         return 0;
2282
2283                 return pagemap_scan_output(categories, p, start, &end);
2284         }
2285
2286         i_mmap_lock_write(vma->vm_file->f_mapping);
2287         ptl = huge_pte_lock(hstate_vma(vma), vma->vm_mm, ptep);
2288
2289         pte = huge_ptep_get(ptep);
2290         categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2291
2292         if (!pagemap_scan_is_interesting_page(categories, p))
2293                 goto out_unlock;
2294
2295         ret = pagemap_scan_output(categories, p, start, &end);
2296         if (start == end)
2297                 goto out_unlock;
2298
2299         if (~categories & PAGE_IS_WRITTEN)
2300                 goto out_unlock;
2301
2302         if (end != start + HPAGE_SIZE) {
2303                 /* Partial HugeTLB page WP isn't possible. */
2304                 pagemap_scan_backout_range(p, start, end);
2305                 p->arg.walk_end = start;
2306                 ret = 0;
2307                 goto out_unlock;
2308         }
2309
2310         make_uffd_wp_huge_pte(vma, start, ptep, pte);
2311         flush_hugetlb_tlb_range(vma, start, end);
2312
2313 out_unlock:
2314         spin_unlock(ptl);
2315         i_mmap_unlock_write(vma->vm_file->f_mapping);
2316
2317         return ret;
2318 }
2319 #else
2320 #define pagemap_scan_hugetlb_entry NULL
2321 #endif
2322
2323 static int pagemap_scan_pte_hole(unsigned long addr, unsigned long end,
2324                                  int depth, struct mm_walk *walk)
2325 {
2326         struct pagemap_scan_private *p = walk->private;
2327         struct vm_area_struct *vma = walk->vma;
2328         int ret, err;
2329
2330         if (!vma || !pagemap_scan_is_interesting_page(p->cur_vma_category, p))
2331                 return 0;
2332
2333         ret = pagemap_scan_output(p->cur_vma_category, p, addr, &end);
2334         if (addr == end)
2335                 return ret;
2336
2337         if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2338                 return ret;
2339
2340         err = uffd_wp_range(vma, addr, end - addr, true);
2341         if (err < 0)
2342                 ret = err;
2343
2344         return ret;
2345 }
2346
2347 static const struct mm_walk_ops pagemap_scan_ops = {
2348         .test_walk = pagemap_scan_test_walk,
2349         .pmd_entry = pagemap_scan_pmd_entry,
2350         .pte_hole = pagemap_scan_pte_hole,
2351         .hugetlb_entry = pagemap_scan_hugetlb_entry,
2352 };
2353
2354 static int pagemap_scan_get_args(struct pm_scan_arg *arg,
2355                                  unsigned long uarg)
2356 {
2357         if (copy_from_user(arg, (void __user *)uarg, sizeof(*arg)))
2358                 return -EFAULT;
2359
2360         if (arg->size != sizeof(struct pm_scan_arg))
2361                 return -EINVAL;
2362
2363         /* Validate requested features */
2364         if (arg->flags & ~PM_SCAN_FLAGS)
2365                 return -EINVAL;
2366         if ((arg->category_inverted | arg->category_mask |
2367              arg->category_anyof_mask | arg->return_mask) & ~PM_SCAN_CATEGORIES)
2368                 return -EINVAL;
2369
2370         arg->start = untagged_addr((unsigned long)arg->start);
2371         arg->end = untagged_addr((unsigned long)arg->end);
2372         arg->vec = untagged_addr((unsigned long)arg->vec);
2373
2374         /* Validate memory pointers */
2375         if (!IS_ALIGNED(arg->start, PAGE_SIZE))
2376                 return -EINVAL;
2377         if (!access_ok((void __user *)(long)arg->start, arg->end - arg->start))
2378                 return -EFAULT;
2379         if (!arg->vec && arg->vec_len)
2380                 return -EINVAL;
2381         if (arg->vec && !access_ok((void __user *)(long)arg->vec,
2382                               arg->vec_len * sizeof(struct page_region)))
2383                 return -EFAULT;
2384
2385         /* Fixup default values */
2386         arg->end = ALIGN(arg->end, PAGE_SIZE);
2387         arg->walk_end = 0;
2388         if (!arg->max_pages)
2389                 arg->max_pages = ULONG_MAX;
2390
2391         return 0;
2392 }
2393
2394 static int pagemap_scan_writeback_args(struct pm_scan_arg *arg,
2395                                        unsigned long uargl)
2396 {
2397         struct pm_scan_arg __user *uarg = (void __user *)uargl;
2398
2399         if (copy_to_user(&uarg->walk_end, &arg->walk_end, sizeof(arg->walk_end)))
2400                 return -EFAULT;
2401
2402         return 0;
2403 }
2404
2405 static int pagemap_scan_init_bounce_buffer(struct pagemap_scan_private *p)
2406 {
2407         if (!p->arg.vec_len)
2408                 return 0;
2409
2410         p->vec_buf_len = min_t(size_t, PAGEMAP_WALK_SIZE >> PAGE_SHIFT,
2411                                p->arg.vec_len);
2412         p->vec_buf = kmalloc_array(p->vec_buf_len, sizeof(*p->vec_buf),
2413                                    GFP_KERNEL);
2414         if (!p->vec_buf)
2415                 return -ENOMEM;
2416
2417         p->vec_buf->start = p->vec_buf->end = 0;
2418         p->vec_out = (struct page_region __user *)(long)p->arg.vec;
2419
2420         return 0;
2421 }
2422
2423 static long pagemap_scan_flush_buffer(struct pagemap_scan_private *p)
2424 {
2425         const struct page_region *buf = p->vec_buf;
2426         long n = p->vec_buf_index;
2427
2428         if (!p->vec_buf)
2429                 return 0;
2430
2431         if (buf[n].end != buf[n].start)
2432                 n++;
2433
2434         if (!n)
2435                 return 0;
2436
2437         if (copy_to_user(p->vec_out, buf, n * sizeof(*buf)))
2438                 return -EFAULT;
2439
2440         p->arg.vec_len -= n;
2441         p->vec_out += n;
2442
2443         p->vec_buf_index = 0;
2444         p->vec_buf_len = min_t(size_t, p->vec_buf_len, p->arg.vec_len);
2445         p->vec_buf->start = p->vec_buf->end = 0;
2446
2447         return n;
2448 }
2449
2450 static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg)
2451 {
2452         struct pagemap_scan_private p = {0};
2453         unsigned long walk_start;
2454         size_t n_ranges_out = 0;
2455         int ret;
2456
2457         ret = pagemap_scan_get_args(&p.arg, uarg);
2458         if (ret)
2459                 return ret;
2460
2461         p.masks_of_interest = p.arg.category_mask | p.arg.category_anyof_mask |
2462                               p.arg.return_mask;
2463         ret = pagemap_scan_init_bounce_buffer(&p);
2464         if (ret)
2465                 return ret;
2466
2467         for (walk_start = p.arg.start; walk_start < p.arg.end;
2468                         walk_start = p.arg.walk_end) {
2469                 struct mmu_notifier_range range;
2470                 long n_out;
2471
2472                 if (fatal_signal_pending(current)) {
2473                         ret = -EINTR;
2474                         break;
2475                 }
2476
2477                 ret = mmap_read_lock_killable(mm);
2478                 if (ret)
2479                         break;
2480
2481                 /* Protection change for the range is going to happen. */
2482                 if (p.arg.flags & PM_SCAN_WP_MATCHING) {
2483                         mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_VMA, 0,
2484                                                 mm, walk_start, p.arg.end);
2485                         mmu_notifier_invalidate_range_start(&range);
2486                 }
2487
2488                 ret = walk_page_range(mm, walk_start, p.arg.end,
2489                                       &pagemap_scan_ops, &p);
2490
2491                 if (p.arg.flags & PM_SCAN_WP_MATCHING)
2492                         mmu_notifier_invalidate_range_end(&range);
2493
2494                 mmap_read_unlock(mm);
2495
2496                 n_out = pagemap_scan_flush_buffer(&p);
2497                 if (n_out < 0)
2498                         ret = n_out;
2499                 else
2500                         n_ranges_out += n_out;
2501
2502                 if (ret != -ENOSPC)
2503                         break;
2504
2505                 if (p.arg.vec_len == 0 || p.found_pages == p.arg.max_pages)
2506                         break;
2507         }
2508
2509         /* ENOSPC signifies early stop (buffer full) from the walk. */
2510         if (!ret || ret == -ENOSPC)
2511                 ret = n_ranges_out;
2512
2513         /* The walk_end isn't set when ret is zero */
2514         if (!p.arg.walk_end)
2515                 p.arg.walk_end = p.arg.end;
2516         if (pagemap_scan_writeback_args(&p.arg, uarg))
2517                 ret = -EFAULT;
2518
2519         kfree(p.vec_buf);
2520         return ret;
2521 }
2522
2523 static long do_pagemap_cmd(struct file *file, unsigned int cmd,
2524                            unsigned long arg)
2525 {
2526         struct mm_struct *mm = file->private_data;
2527
2528         switch (cmd) {
2529         case PAGEMAP_SCAN:
2530                 return do_pagemap_scan(mm, arg);
2531
2532         default:
2533                 return -EINVAL;
2534         }
2535 }
2536
2537 const struct file_operations proc_pagemap_operations = {
2538         .llseek         = mem_lseek, /* borrow this */
2539         .read           = pagemap_read,
2540         .open           = pagemap_open,
2541         .release        = pagemap_release,
2542         .unlocked_ioctl = do_pagemap_cmd,
2543         .compat_ioctl   = do_pagemap_cmd,
2544 };
2545 #endif /* CONFIG_PROC_PAGE_MONITOR */
2546
2547 #ifdef CONFIG_NUMA
2548
2549 struct numa_maps {
2550         unsigned long pages;
2551         unsigned long anon;
2552         unsigned long active;
2553         unsigned long writeback;
2554         unsigned long mapcount_max;
2555         unsigned long dirty;
2556         unsigned long swapcache;
2557         unsigned long node[MAX_NUMNODES];
2558 };
2559
2560 struct numa_maps_private {
2561         struct proc_maps_private proc_maps;
2562         struct numa_maps md;
2563 };
2564
2565 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
2566                         unsigned long nr_pages)
2567 {
2568         struct folio *folio = page_folio(page);
2569         int count = page_mapcount(page);
2570
2571         md->pages += nr_pages;
2572         if (pte_dirty || folio_test_dirty(folio))
2573                 md->dirty += nr_pages;
2574
2575         if (folio_test_swapcache(folio))
2576                 md->swapcache += nr_pages;
2577
2578         if (folio_test_active(folio) || folio_test_unevictable(folio))
2579                 md->active += nr_pages;
2580
2581         if (folio_test_writeback(folio))
2582                 md->writeback += nr_pages;
2583
2584         if (folio_test_anon(folio))
2585                 md->anon += nr_pages;
2586
2587         if (count > md->mapcount_max)
2588                 md->mapcount_max = count;
2589
2590         md->node[folio_nid(folio)] += nr_pages;
2591 }
2592
2593 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
2594                 unsigned long addr)
2595 {
2596         struct page *page;
2597         int nid;
2598
2599         if (!pte_present(pte))
2600                 return NULL;
2601
2602         page = vm_normal_page(vma, addr, pte);
2603         if (!page || is_zone_device_page(page))
2604                 return NULL;
2605
2606         if (PageReserved(page))
2607                 return NULL;
2608
2609         nid = page_to_nid(page);
2610         if (!node_isset(nid, node_states[N_MEMORY]))
2611                 return NULL;
2612
2613         return page;
2614 }
2615
2616 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2617 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
2618                                               struct vm_area_struct *vma,
2619                                               unsigned long addr)
2620 {
2621         struct page *page;
2622         int nid;
2623
2624         if (!pmd_present(pmd))
2625                 return NULL;
2626
2627         page = vm_normal_page_pmd(vma, addr, pmd);
2628         if (!page)
2629                 return NULL;
2630
2631         if (PageReserved(page))
2632                 return NULL;
2633
2634         nid = page_to_nid(page);
2635         if (!node_isset(nid, node_states[N_MEMORY]))
2636                 return NULL;
2637
2638         return page;
2639 }
2640 #endif
2641
2642 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
2643                 unsigned long end, struct mm_walk *walk)
2644 {
2645         struct numa_maps *md = walk->private;
2646         struct vm_area_struct *vma = walk->vma;
2647         spinlock_t *ptl;
2648         pte_t *orig_pte;
2649         pte_t *pte;
2650
2651 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2652         ptl = pmd_trans_huge_lock(pmd, vma);
2653         if (ptl) {
2654                 struct page *page;
2655
2656                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
2657                 if (page)
2658                         gather_stats(page, md, pmd_dirty(*pmd),
2659                                      HPAGE_PMD_SIZE/PAGE_SIZE);
2660                 spin_unlock(ptl);
2661                 return 0;
2662         }
2663 #endif
2664         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
2665         if (!pte) {
2666                 walk->action = ACTION_AGAIN;
2667                 return 0;
2668         }
2669         do {
2670                 pte_t ptent = ptep_get(pte);
2671                 struct page *page = can_gather_numa_stats(ptent, vma, addr);
2672                 if (!page)
2673                         continue;
2674                 gather_stats(page, md, pte_dirty(ptent), 1);
2675
2676         } while (pte++, addr += PAGE_SIZE, addr != end);
2677         pte_unmap_unlock(orig_pte, ptl);
2678         cond_resched();
2679         return 0;
2680 }
2681 #ifdef CONFIG_HUGETLB_PAGE
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         pte_t huge_pte = huge_ptep_get(pte);
2686         struct numa_maps *md;
2687         struct page *page;
2688
2689         if (!pte_present(huge_pte))
2690                 return 0;
2691
2692         page = pte_page(huge_pte);
2693
2694         md = walk->private;
2695         gather_stats(page, md, pte_dirty(huge_pte), 1);
2696         return 0;
2697 }
2698
2699 #else
2700 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
2701                 unsigned long addr, unsigned long end, struct mm_walk *walk)
2702 {
2703         return 0;
2704 }
2705 #endif
2706
2707 static const struct mm_walk_ops show_numa_ops = {
2708         .hugetlb_entry = gather_hugetlb_stats,
2709         .pmd_entry = gather_pte_stats,
2710         .walk_lock = PGWALK_RDLOCK,
2711 };
2712
2713 /*
2714  * Display pages allocated per node and memory policy via /proc.
2715  */
2716 static int show_numa_map(struct seq_file *m, void *v)
2717 {
2718         struct numa_maps_private *numa_priv = m->private;
2719         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
2720         struct vm_area_struct *vma = v;
2721         struct numa_maps *md = &numa_priv->md;
2722         struct file *file = vma->vm_file;
2723         struct mm_struct *mm = vma->vm_mm;
2724         char buffer[64];
2725         struct mempolicy *pol;
2726         pgoff_t ilx;
2727         int nid;
2728
2729         if (!mm)
2730                 return 0;
2731
2732         /* Ensure we start with an empty set of numa_maps statistics. */
2733         memset(md, 0, sizeof(*md));
2734
2735         pol = __get_vma_policy(vma, vma->vm_start, &ilx);
2736         if (pol) {
2737                 mpol_to_str(buffer, sizeof(buffer), pol);
2738                 mpol_cond_put(pol);
2739         } else {
2740                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
2741         }
2742
2743         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
2744
2745         if (file) {
2746                 seq_puts(m, " file=");
2747                 seq_path(m, file_user_path(file), "\n\t= ");
2748         } else if (vma_is_initial_heap(vma)) {
2749                 seq_puts(m, " heap");
2750         } else if (vma_is_initial_stack(vma)) {
2751                 seq_puts(m, " stack");
2752         }
2753
2754         if (is_vm_hugetlb_page(vma))
2755                 seq_puts(m, " huge");
2756
2757         /* mmap_lock is held by m_start */
2758         walk_page_vma(vma, &show_numa_ops, md);
2759
2760         if (!md->pages)
2761                 goto out;
2762
2763         if (md->anon)
2764                 seq_printf(m, " anon=%lu", md->anon);
2765
2766         if (md->dirty)
2767                 seq_printf(m, " dirty=%lu", md->dirty);
2768
2769         if (md->pages != md->anon && md->pages != md->dirty)
2770                 seq_printf(m, " mapped=%lu", md->pages);
2771
2772         if (md->mapcount_max > 1)
2773                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
2774
2775         if (md->swapcache)
2776                 seq_printf(m, " swapcache=%lu", md->swapcache);
2777
2778         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2779                 seq_printf(m, " active=%lu", md->active);
2780
2781         if (md->writeback)
2782                 seq_printf(m, " writeback=%lu", md->writeback);
2783
2784         for_each_node_state(nid, N_MEMORY)
2785                 if (md->node[nid])
2786                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2787
2788         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2789 out:
2790         seq_putc(m, '\n');
2791         return 0;
2792 }
2793
2794 static const struct seq_operations proc_pid_numa_maps_op = {
2795         .start  = m_start,
2796         .next   = m_next,
2797         .stop   = m_stop,
2798         .show   = show_numa_map,
2799 };
2800
2801 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2802 {
2803         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2804                                 sizeof(struct numa_maps_private));
2805 }
2806
2807 const struct file_operations proc_pid_numa_maps_operations = {
2808         .open           = pid_numa_maps_open,
2809         .read           = seq_read,
2810         .llseek         = seq_lseek,
2811         .release        = proc_map_release,
2812 };
2813
2814 #endif /* CONFIG_NUMA */