fs/proc/task_mmu.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/pagewalk.h>
   3 #include <linux/mm_inline.h>
   4 #include <linux/hugetlb.h>
   5 #include <linux/huge_mm.h>
   6 #include <linux/mount.h>
   7 #include <linux/ksm.h>
   8 #include <linux/seq_file.h>
   9 #include <linux/highmem.h>
  10 #include <linux/ptrace.h>
  11 #include <linux/slab.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/mempolicy.h>
  14 #include <linux/rmap.h>
  15 #include <linux/swap.h>
  16 #include <linux/sched/mm.h>
  17 #include <linux/swapops.h>
  18 #include <linux/mmu_notifier.h>
  19 #include <linux/page_idle.h>
  20 #include <linux/shmem_fs.h>
  21 #include <linux/uaccess.h>
  22 #include <linux/pkeys.h>
  23 #include <linux/minmax.h>
  24 #include <linux/overflow.h>
  25
  26 #include <asm/elf.h>
  27 #include <asm/tlb.h>
  28 #include <asm/tlbflush.h>
  29 #include "internal.h"
  30
  31 #define SEQ_PUT_DEC(str, val) \
  32                 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
  33 void task_mem(struct seq_file *m, struct mm_struct *mm)
  34 {
  35         unsigned long text, lib, swap, anon, file, shmem;
  36         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  37
  38         anon = get_mm_counter(mm, MM_ANONPAGES);
  39         file = get_mm_counter(mm, MM_FILEPAGES);
  40         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  41
  42         /*
  43          * Note: to minimize their overhead, mm maintains hiwater_vm and
  44          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  45          * collector of these hiwater stats must therefore get total_vm
  46          * and rss too, which will usually be the higher.  Barriers? not
  47          * worth the effort, such snapshots can always be inconsistent.
  48          */
  49         hiwater_vm = total_vm = mm->total_vm;
  50         if (hiwater_vm < mm->hiwater_vm)
  51                 hiwater_vm = mm->hiwater_vm;
  52         hiwater_rss = total_rss = anon + file + shmem;
  53         if (hiwater_rss < mm->hiwater_rss)
  54                 hiwater_rss = mm->hiwater_rss;
  55
  56         /* split executable areas between text and lib */
  57         text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
  58         text = min(text, mm->exec_vm << PAGE_SHIFT);
  59         lib = (mm->exec_vm << PAGE_SHIFT) - text;
  60
  61         swap = get_mm_counter(mm, MM_SWAPENTS);
  62         SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
  63         SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
  64         SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
  65         SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
  66         SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
  67         SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
  68         SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
  69         SEQ_PUT_DEC(" kB\nRssFile:\t", file);
  70         SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
  71         SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
  72         SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
  73         seq_put_decimal_ull_width(m,
  74                     " kB\nVmExe:\t", text >> 10, 8);
  75         seq_put_decimal_ull_width(m,
  76                     " kB\nVmLib:\t", lib >> 10, 8);
  77         seq_put_decimal_ull_width(m,
  78                     " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
  79         SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
  80         seq_puts(m, " kB\n");
  81         hugetlb_report_usage(m, mm);
  82 }
  83 #undef SEQ_PUT_DEC
  84
  85 unsigned long task_vsize(struct mm_struct *mm)
  86 {
  87         return PAGE_SIZE * mm->total_vm;
  88 }
  89
  90 unsigned long task_statm(struct mm_struct *mm,
  91                          unsigned long *shared, unsigned long *text,
  92                          unsigned long *data, unsigned long *resident)
  93 {
  94         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  95                         get_mm_counter(mm, MM_SHMEMPAGES);
  96         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  97                                                                 >> PAGE_SHIFT;
  98         *data = mm->data_vm + mm->stack_vm;
  99         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
 100         return mm->total_vm;
 101 }
 102
 103 #ifdef CONFIG_NUMA
 104 /*
 105  * Save get_task_policy() for show_numa_map().
 106  */
 107 static void hold_task_mempolicy(struct proc_maps_private *priv)
 108 {
 109         struct task_struct *task = priv->task;
 110
 111         task_lock(task);
 112         priv->task_mempolicy = get_task_policy(task);
 113         mpol_get(priv->task_mempolicy);
 114         task_unlock(task);
 115 }
 116 static void release_task_mempolicy(struct proc_maps_private *priv)
 117 {
 118         mpol_put(priv->task_mempolicy);
 119 }
 120 #else
 121 static void hold_task_mempolicy(struct proc_maps_private *priv)
 122 {
 123 }
 124 static void release_task_mempolicy(struct proc_maps_private *priv)
 125 {
 126 }
 127 #endif
 128
 129 static struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv,
 130                                                 loff_t *ppos)
 131 {
 132         struct vm_area_struct *vma = vma_next(&priv->iter);
 133
 134         if (vma) {
 135                 *ppos = vma->vm_start;
 136         } else {
 137                 *ppos = -2UL;
 138                 vma = get_gate_vma(priv->mm);
 139         }
 140
 141         return vma;
 142 }
 143
 144 static void *m_start(struct seq_file *m, loff_t *ppos)
 145 {
 146         struct proc_maps_private *priv = m->private;
 147         unsigned long last_addr = *ppos;
 148         struct mm_struct *mm;
 149
 150         /* See m_next(). Zero at the start or after lseek. */
 151         if (last_addr == -1UL)
 152                 return NULL;
 153
 154         priv->task = get_proc_task(priv->inode);
 155         if (!priv->task)
 156                 return ERR_PTR(-ESRCH);
 157
 158         mm = priv->mm;
 159         if (!mm || !mmget_not_zero(mm)) {
 160                 put_task_struct(priv->task);
 161                 priv->task = NULL;
 162                 return NULL;
 163         }
 164
 165         if (mmap_read_lock_killable(mm)) {
 166                 mmput(mm);
 167                 put_task_struct(priv->task);
 168                 priv->task = NULL;
 169                 return ERR_PTR(-EINTR);
 170         }
 171
 172         vma_iter_init(&priv->iter, mm, last_addr);
 173         hold_task_mempolicy(priv);
 174         if (last_addr == -2UL)
 175                 return get_gate_vma(mm);
 176
 177         return proc_get_vma(priv, ppos);
 178 }
 179
 180 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
 181 {
 182         if (*ppos == -2UL) {
 183                 *ppos = -1UL;
 184                 return NULL;
 185         }
 186         return proc_get_vma(m->private, ppos);
 187 }
 188
 189 static void m_stop(struct seq_file *m, void *v)
 190 {
 191         struct proc_maps_private *priv = m->private;
 192         struct mm_struct *mm = priv->mm;
 193
 194         if (!priv->task)
 195                 return;
 196
 197         release_task_mempolicy(priv);
 198         mmap_read_unlock(mm);
 199         mmput(mm);
 200         put_task_struct(priv->task);
 201         priv->task = NULL;
 202 }
 203
 204 static int proc_maps_open(struct inode *inode, struct file *file,
 205                         const struct seq_operations *ops, int psize)
 206 {
 207         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 208
 209         if (!priv)
 210                 return -ENOMEM;
 211
 212         priv->inode = inode;
 213         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 214         if (IS_ERR(priv->mm)) {
 215                 int err = PTR_ERR(priv->mm);
 216
 217                 seq_release_private(inode, file);
 218                 return err;
 219         }
 220
 221         return 0;
 222 }
 223
 224 static int proc_map_release(struct inode *inode, struct file *file)
 225 {
 226         struct seq_file *seq = file->private_data;
 227         struct proc_maps_private *priv = seq->private;
 228
 229         if (priv->mm)
 230                 mmdrop(priv->mm);
 231
 232         return seq_release_private(inode, file);
 233 }
 234
 235 static int do_maps_open(struct inode *inode, struct file *file,
 236                         const struct seq_operations *ops)
 237 {
 238         return proc_maps_open(inode, file, ops,
 239                                 sizeof(struct proc_maps_private));
 240 }
 241
 242 static void show_vma_header_prefix(struct seq_file *m,
 243                                    unsigned long start, unsigned long end,
 244                                    vm_flags_t flags, unsigned long long pgoff,
 245                                    dev_t dev, unsigned long ino)
 246 {
 247         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 248         seq_put_hex_ll(m, NULL, start, 8);
 249         seq_put_hex_ll(m, "-", end, 8);
 250         seq_putc(m, ' ');
 251         seq_putc(m, flags & VM_READ ? 'r' : '-');
 252         seq_putc(m, flags & VM_WRITE ? 'w' : '-');
 253         seq_putc(m, flags & VM_EXEC ? 'x' : '-');
 254         seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
 255         seq_put_hex_ll(m, " ", pgoff, 8);
 256         seq_put_hex_ll(m, " ", MAJOR(dev), 2);
 257         seq_put_hex_ll(m, ":", MINOR(dev), 2);
 258         seq_put_decimal_ull(m, " ", ino);
 259         seq_putc(m, ' ');
 260 }
 261
 262 static void
 263 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 264 {
 265         struct anon_vma_name *anon_name = NULL;
 266         struct mm_struct *mm = vma->vm_mm;
 267         struct file *file = vma->vm_file;
 268         vm_flags_t flags = vma->vm_flags;
 269         unsigned long ino = 0;
 270         unsigned long long pgoff = 0;
 271         unsigned long start, end;
 272         dev_t dev = 0;
 273         const char *name = NULL;
 274
 275         if (file) {
 276                 struct inode *inode = file_inode(vma->vm_file);
 277                 dev = inode->i_sb->s_dev;
 278                 ino = inode->i_ino;
 279                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 280         }
 281
 282         start = vma->vm_start;
 283         end = vma->vm_end;
 284         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 285         if (mm)
 286                 anon_name = anon_vma_name(vma);
 287
 288         /*
 289          * Print the dentry name for named mappings, and a
 290          * special [heap] marker for the heap:
 291          */
 292         if (file) {
 293                 seq_pad(m, ' ');
 294                 /*
 295                  * If user named this anon shared memory via
 296                  * prctl(PR_SET_VMA ..., use the provided name.
 297                  */
 298                 if (anon_name)
 299                         seq_printf(m, "[anon_shmem:%s]", anon_name->name);
 300                 else
 301                         seq_path(m, file_user_path(file), "\n");
 302                 goto done;
 303         }
 304
 305         if (vma->vm_ops && vma->vm_ops->name) {
 306                 name = vma->vm_ops->name(vma);
 307                 if (name)
 308                         goto done;
 309         }
 310
 311         name = arch_vma_name(vma);
 312         if (!name) {
 313                 if (!mm) {
 314                         name = "[vdso]";
 315                         goto done;
 316                 }
 317
 318                 if (vma_is_initial_heap(vma)) {
 319                         name = "[heap]";
 320                         goto done;
 321                 }
 322
 323                 if (vma_is_initial_stack(vma)) {
 324                         name = "[stack]";
 325                         goto done;
 326                 }
 327
 328                 if (anon_name) {
 329                         seq_pad(m, ' ');
 330                         seq_printf(m, "[anon:%s]", anon_name->name);
 331                 }
 332         }
 333
 334 done:
 335         if (name) {
 336                 seq_pad(m, ' ');
 337                 seq_puts(m, name);
 338         }
 339         seq_putc(m, '\n');
 340 }
 341
 342 static int show_map(struct seq_file *m, void *v)
 343 {
 344         show_map_vma(m, v);
 345         return 0;
 346 }
 347
 348 static const struct seq_operations proc_pid_maps_op = {
 349         .start  = m_start,
 350         .next   = m_next,
 351         .stop   = m_stop,
 352         .show   = show_map
 353 };
 354
 355 static int pid_maps_open(struct inode *inode, struct file *file)
 356 {
 357         return do_maps_open(inode, file, &proc_pid_maps_op);
 358 }
 359
 360 const struct file_operations proc_pid_maps_operations = {
 361         .open           = pid_maps_open,
 362         .read           = seq_read,
 363         .llseek         = seq_lseek,
 364         .release        = proc_map_release,
 365 };
 366
 367 /*
 368  * Proportional Set Size(PSS): my share of RSS.
 369  *
 370  * PSS of a process is the count of pages it has in memory, where each
 371  * page is divided by the number of processes sharing it.  So if a
 372  * process has 1000 pages all to itself, and 1000 shared with one other
 373  * process, its PSS will be 1500.
 374  *
 375  * To keep (accumulated) division errors low, we adopt a 64bit
 376  * fixed-point pss counter to minimize division errors. So (pss >>
 377  * PSS_SHIFT) would be the real byte count.
 378  *
 379  * A shift of 12 before division means (assuming 4K page size):
 380  *      - 1M 3-user-pages add up to 8KB errors;
 381  *      - supports mapcount up to 2^24, or 16M;
 382  *      - supports PSS up to 2^52 bytes, or 4PB.
 383  */
 384 #define PSS_SHIFT 12
 385
 386 #ifdef CONFIG_PROC_PAGE_MONITOR
 387 struct mem_size_stats {
 388         unsigned long resident;
 389         unsigned long shared_clean;
 390         unsigned long shared_dirty;
 391         unsigned long private_clean;
 392         unsigned long private_dirty;
 393         unsigned long referenced;
 394         unsigned long anonymous;
 395         unsigned long lazyfree;
 396         unsigned long anonymous_thp;
 397         unsigned long shmem_thp;
 398         unsigned long file_thp;
 399         unsigned long swap;
 400         unsigned long shared_hugetlb;
 401         unsigned long private_hugetlb;
 402         unsigned long ksm;
 403         u64 pss;
 404         u64 pss_anon;
 405         u64 pss_file;
 406         u64 pss_shmem;
 407         u64 pss_dirty;
 408         u64 pss_locked;
 409         u64 swap_pss;
 410 };
 411
 412 static void smaps_page_accumulate(struct mem_size_stats *mss,
 413                 struct page *page, unsigned long size, unsigned long pss,
 414                 bool dirty, bool locked, bool private)
 415 {
 416         mss->pss += pss;
 417
 418         if (PageAnon(page))
 419                 mss->pss_anon += pss;
 420         else if (PageSwapBacked(page))
 421                 mss->pss_shmem += pss;
 422         else
 423                 mss->pss_file += pss;
 424
 425         if (locked)
 426                 mss->pss_locked += pss;
 427
 428         if (dirty || PageDirty(page)) {
 429                 mss->pss_dirty += pss;
 430                 if (private)
 431                         mss->private_dirty += size;
 432                 else
 433                         mss->shared_dirty += size;
 434         } else {
 435                 if (private)
 436                         mss->private_clean += size;
 437                 else
 438                         mss->shared_clean += size;
 439         }
 440 }
 441
 442 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 443                 bool compound, bool young, bool dirty, bool locked,
 444                 bool migration)
 445 {
 446         int i, nr = compound ? compound_nr(page) : 1;
 447         unsigned long size = nr * PAGE_SIZE;
 448
 449         /*
 450          * First accumulate quantities that depend only on |size| and the type
 451          * of the compound page.
 452          */
 453         if (PageAnon(page)) {
 454                 mss->anonymous += size;
 455                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 456                         mss->lazyfree += size;
 457         }
 458
 459         if (PageKsm(page))
 460                 mss->ksm += size;
 461
 462         mss->resident += size;
 463         /* Accumulate the size in pages that have been accessed. */
 464         if (young || page_is_young(page) || PageReferenced(page))
 465                 mss->referenced += size;
 466
 467         /*
 468          * Then accumulate quantities that may depend on sharing, or that may
 469          * differ page-by-page.
 470          *
 471          * page_count(page) == 1 guarantees the page is mapped exactly once.
 472          * If any subpage of the compound page mapped with PTE it would elevate
 473          * page_count().
 474          *
 475          * The page_mapcount() is called to get a snapshot of the mapcount.
 476          * Without holding the page lock this snapshot can be slightly wrong as
 477          * we cannot always read the mapcount atomically.  It is not safe to
 478          * call page_mapcount() even with PTL held if the page is not mapped,
 479          * especially for migration entries.  Treat regular migration entries
 480          * as mapcount == 1.
 481          */
 482         if ((page_count(page) == 1) || migration) {
 483                 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
 484                         locked, true);
 485                 return;
 486         }
 487         for (i = 0; i < nr; i++, page++) {
 488                 int mapcount = page_mapcount(page);
 489                 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
 490                 if (mapcount >= 2)
 491                         pss /= mapcount;
 492                 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
 493                                       mapcount < 2);
 494         }
 495 }
 496
 497 #ifdef CONFIG_SHMEM
 498 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 499                           __always_unused int depth, struct mm_walk *walk)
 500 {
 501         struct mem_size_stats *mss = walk->private;
 502         struct vm_area_struct *vma = walk->vma;
 503
 504         mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping,
 505                                               linear_page_index(vma, addr),
 506                                               linear_page_index(vma, end));
 507
 508         return 0;
 509 }
 510 #else
 511 #define smaps_pte_hole          NULL
 512 #endif /* CONFIG_SHMEM */
 513
 514 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
 515 {
 516 #ifdef CONFIG_SHMEM
 517         if (walk->ops->pte_hole) {
 518                 /* depth is not used */
 519                 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
 520         }
 521 #endif
 522 }
 523
 524 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 525                 struct mm_walk *walk)
 526 {
 527         struct mem_size_stats *mss = walk->private;
 528         struct vm_area_struct *vma = walk->vma;
 529         bool locked = !!(vma->vm_flags & VM_LOCKED);
 530         struct page *page = NULL;
 531         bool migration = false, young = false, dirty = false;
 532         pte_t ptent = ptep_get(pte);
 533
 534         if (pte_present(ptent)) {
 535                 page = vm_normal_page(vma, addr, ptent);
 536                 young = pte_young(ptent);
 537                 dirty = pte_dirty(ptent);
 538         } else if (is_swap_pte(ptent)) {
 539                 swp_entry_t swpent = pte_to_swp_entry(ptent);
 540
 541                 if (!non_swap_entry(swpent)) {
 542                         int mapcount;
 543
 544                         mss->swap += PAGE_SIZE;
 545                         mapcount = swp_swapcount(swpent);
 546                         if (mapcount >= 2) {
 547                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 548
 549                                 do_div(pss_delta, mapcount);
 550                                 mss->swap_pss += pss_delta;
 551                         } else {
 552                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 553                         }
 554                 } else if (is_pfn_swap_entry(swpent)) {
 555                         if (is_migration_entry(swpent))
 556                                 migration = true;
 557                         page = pfn_swap_entry_to_page(swpent);
 558                 }
 559         } else {
 560                 smaps_pte_hole_lookup(addr, walk);
 561                 return;
 562         }
 563
 564         if (!page)
 565                 return;
 566
 567         smaps_account(mss, page, false, young, dirty, locked, migration);
 568 }
 569
 570 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 571 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 572                 struct mm_walk *walk)
 573 {
 574         struct mem_size_stats *mss = walk->private;
 575         struct vm_area_struct *vma = walk->vma;
 576         bool locked = !!(vma->vm_flags & VM_LOCKED);
 577         struct page *page = NULL;
 578         bool migration = false;
 579
 580         if (pmd_present(*pmd)) {
 581                 page = vm_normal_page_pmd(vma, addr, *pmd);
 582         } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
 583                 swp_entry_t entry = pmd_to_swp_entry(*pmd);
 584
 585                 if (is_migration_entry(entry)) {
 586                         migration = true;
 587                         page = pfn_swap_entry_to_page(entry);
 588                 }
 589         }
 590         if (IS_ERR_OR_NULL(page))
 591                 return;
 592         if (PageAnon(page))
 593                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 594         else if (PageSwapBacked(page))
 595                 mss->shmem_thp += HPAGE_PMD_SIZE;
 596         else if (is_zone_device_page(page))
 597                 /* pass */;
 598         else
 599                 mss->file_thp += HPAGE_PMD_SIZE;
 600
 601         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
 602                       locked, migration);
 603 }
 604 #else
 605 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 606                 struct mm_walk *walk)
 607 {
 608 }
 609 #endif
 610
 611 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 612                            struct mm_walk *walk)
 613 {
 614         struct vm_area_struct *vma = walk->vma;
 615         pte_t *pte;
 616         spinlock_t *ptl;
 617
 618         ptl = pmd_trans_huge_lock(pmd, vma);
 619         if (ptl) {
 620                 smaps_pmd_entry(pmd, addr, walk);
 621                 spin_unlock(ptl);
 622                 goto out;
 623         }
 624
 625         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 626         if (!pte) {
 627                 walk->action = ACTION_AGAIN;
 628                 return 0;
 629         }
 630         for (; addr != end; pte++, addr += PAGE_SIZE)
 631                 smaps_pte_entry(pte, addr, walk);
 632         pte_unmap_unlock(pte - 1, ptl);
 633 out:
 634         cond_resched();
 635         return 0;
 636 }
 637
 638 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 639 {
 640         /*
 641          * Don't forget to update Documentation/ on changes.
 642          */
 643         static const char mnemonics[BITS_PER_LONG][2] = {
 644                 /*
 645                  * In case if we meet a flag we don't know about.
 646                  */
 647                 [0 ... (BITS_PER_LONG-1)] = "??",
 648
 649                 [ilog2(VM_READ)]        = "rd",
 650                 [ilog2(VM_WRITE)]       = "wr",
 651                 [ilog2(VM_EXEC)]        = "ex",
 652                 [ilog2(VM_SHARED)]      = "sh",
 653                 [ilog2(VM_MAYREAD)]     = "mr",
 654                 [ilog2(VM_MAYWRITE)]    = "mw",
 655                 [ilog2(VM_MAYEXEC)]     = "me",
 656                 [ilog2(VM_MAYSHARE)]    = "ms",
 657                 [ilog2(VM_GROWSDOWN)]   = "gd",
 658                 [ilog2(VM_PFNMAP)]      = "pf",
 659                 [ilog2(VM_LOCKED)]      = "lo",
 660                 [ilog2(VM_IO)]          = "io",
 661                 [ilog2(VM_SEQ_READ)]    = "sr",
 662                 [ilog2(VM_RAND_READ)]   = "rr",
 663                 [ilog2(VM_DONTCOPY)]    = "dc",
 664                 [ilog2(VM_DONTEXPAND)]  = "de",
 665                 [ilog2(VM_LOCKONFAULT)] = "lf",
 666                 [ilog2(VM_ACCOUNT)]     = "ac",
 667                 [ilog2(VM_NORESERVE)]   = "nr",
 668                 [ilog2(VM_HUGETLB)]     = "ht",
 669                 [ilog2(VM_SYNC)]        = "sf",
 670                 [ilog2(VM_ARCH_1)]      = "ar",
 671                 [ilog2(VM_WIPEONFORK)]  = "wf",
 672                 [ilog2(VM_DONTDUMP)]    = "dd",
 673 #ifdef CONFIG_ARM64_BTI
 674                 [ilog2(VM_ARM64_BTI)]   = "bt",
 675 #endif
 676 #ifdef CONFIG_MEM_SOFT_DIRTY
 677                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 678 #endif
 679                 [ilog2(VM_MIXEDMAP)]    = "mm",
 680                 [ilog2(VM_HUGEPAGE)]    = "hg",
 681                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 682                 [ilog2(VM_MERGEABLE)]   = "mg",
 683                 [ilog2(VM_UFFD_MISSING)]= "um",
 684                 [ilog2(VM_UFFD_WP)]     = "uw",
 685 #ifdef CONFIG_ARM64_MTE
 686                 [ilog2(VM_MTE)]         = "mt",
 687                 [ilog2(VM_MTE_ALLOWED)] = "",
 688 #endif
 689 #ifdef CONFIG_ARCH_HAS_PKEYS
 690                 /* These come out via ProtectionKey: */
 691                 [ilog2(VM_PKEY_BIT0)]   = "",
 692                 [ilog2(VM_PKEY_BIT1)]   = "",
 693                 [ilog2(VM_PKEY_BIT2)]   = "",
 694                 [ilog2(VM_PKEY_BIT3)]   = "",
 695 #if VM_PKEY_BIT4
 696                 [ilog2(VM_PKEY_BIT4)]   = "",
 697 #endif
 698 #endif /* CONFIG_ARCH_HAS_PKEYS */
 699 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
 700                 [ilog2(VM_UFFD_MINOR)]  = "ui",
 701 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
 702 #ifdef CONFIG_X86_USER_SHADOW_STACK
 703                 [ilog2(VM_SHADOW_STACK)] = "ss",
 704 #endif
 705         };
 706         size_t i;
 707
 708         seq_puts(m, "VmFlags: ");
 709         for (i = 0; i < BITS_PER_LONG; i++) {
 710                 if (!mnemonics[i][0])
 711                         continue;
 712                 if (vma->vm_flags & (1UL << i)) {
 713                         seq_putc(m, mnemonics[i][0]);
 714                         seq_putc(m, mnemonics[i][1]);
 715                         seq_putc(m, ' ');
 716                 }
 717         }
 718         seq_putc(m, '\n');
 719 }
 720
 721 #ifdef CONFIG_HUGETLB_PAGE
 722 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 723                                  unsigned long addr, unsigned long end,
 724                                  struct mm_walk *walk)
 725 {
 726         struct mem_size_stats *mss = walk->private;
 727         struct vm_area_struct *vma = walk->vma;
 728         struct page *page = NULL;
 729         pte_t ptent = ptep_get(pte);
 730
 731         if (pte_present(ptent)) {
 732                 page = vm_normal_page(vma, addr, ptent);
 733         } else if (is_swap_pte(ptent)) {
 734                 swp_entry_t swpent = pte_to_swp_entry(ptent);
 735
 736                 if (is_pfn_swap_entry(swpent))
 737                         page = pfn_swap_entry_to_page(swpent);
 738         }
 739         if (page) {
 740                 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
 741                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 742                 else
 743                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 744         }
 745         return 0;
 746 }
 747 #else
 748 #define smaps_hugetlb_range     NULL
 749 #endif /* HUGETLB_PAGE */
 750
 751 static const struct mm_walk_ops smaps_walk_ops = {
 752         .pmd_entry              = smaps_pte_range,
 753         .hugetlb_entry          = smaps_hugetlb_range,
 754         .walk_lock              = PGWALK_RDLOCK,
 755 };
 756
 757 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 758         .pmd_entry              = smaps_pte_range,
 759         .hugetlb_entry          = smaps_hugetlb_range,
 760         .pte_hole               = smaps_pte_hole,
 761         .walk_lock              = PGWALK_RDLOCK,
 762 };
 763
 764 /*
 765  * Gather mem stats from @vma with the indicated beginning
 766  * address @start, and keep them in @mss.
 767  *
 768  * Use vm_start of @vma as the beginning address if @start is 0.
 769  */
 770 static void smap_gather_stats(struct vm_area_struct *vma,
 771                 struct mem_size_stats *mss, unsigned long start)
 772 {
 773         const struct mm_walk_ops *ops = &smaps_walk_ops;
 774
 775         /* Invalid start */
 776         if (start >= vma->vm_end)
 777                 return;
 778
 779         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 780                 /*
 781                  * For shared or readonly shmem mappings we know that all
 782                  * swapped out pages belong to the shmem object, and we can
 783                  * obtain the swap value much more efficiently. For private
 784                  * writable mappings, we might have COW pages that are
 785                  * not affected by the parent swapped out pages of the shmem
 786                  * object, so we have to distinguish them during the page walk.
 787                  * Unless we know that the shmem object (or the part mapped by
 788                  * our VMA) has no swapped out pages at all.
 789                  */
 790                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 791
 792                 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 793                                         !(vma->vm_flags & VM_WRITE))) {
 794                         mss->swap += shmem_swapped;
 795                 } else {
 796                         ops = &smaps_shmem_walk_ops;
 797                 }
 798         }
 799
 800         /* mmap_lock is held in m_start */
 801         if (!start)
 802                 walk_page_vma(vma, ops, mss);
 803         else
 804                 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
 805 }
 806
 807 #define SEQ_PUT_DEC(str, val) \
 808                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 809
 810 /* Show the contents common for smaps and smaps_rollup */
 811 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 812         bool rollup_mode)
 813 {
 814         SEQ_PUT_DEC("Rss:            ", mss->resident);
 815         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 816         SEQ_PUT_DEC(" kB\nPss_Dirty:      ", mss->pss_dirty >> PSS_SHIFT);
 817         if (rollup_mode) {
 818                 /*
 819                  * These are meaningful only for smaps_rollup, otherwise two of
 820                  * them are zero, and the other one is the same as Pss.
 821                  */
 822                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 823                         mss->pss_anon >> PSS_SHIFT);
 824                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 825                         mss->pss_file >> PSS_SHIFT);
 826                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 827                         mss->pss_shmem >> PSS_SHIFT);
 828         }
 829         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 830         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 831         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 832         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 833         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 834         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 835         SEQ_PUT_DEC(" kB\nKSM:            ", mss->ksm);
 836         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 837         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 838         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 839         SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
 840         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 841         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 842                                   mss->private_hugetlb >> 10, 7);
 843         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 844         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 845                                         mss->swap_pss >> PSS_SHIFT);
 846         SEQ_PUT_DEC(" kB\nLocked:         ",
 847                                         mss->pss_locked >> PSS_SHIFT);
 848         seq_puts(m, " kB\n");
 849 }
 850
 851 static int show_smap(struct seq_file *m, void *v)
 852 {
 853         struct vm_area_struct *vma = v;
 854         struct mem_size_stats mss = {};
 855
 856         smap_gather_stats(vma, &mss, 0);
 857
 858         show_map_vma(m, vma);
 859
 860         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 861         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 862         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 863         seq_puts(m, " kB\n");
 864
 865         __show_smap(m, &mss, false);
 866
 867         seq_printf(m, "THPeligible:    %8u\n",
 868                    hugepage_vma_check(vma, vma->vm_flags, true, false, true));
 869
 870         if (arch_pkeys_enabled())
 871                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 872         show_smap_vma_flags(m, vma);
 873
 874         return 0;
 875 }
 876
 877 static int show_smaps_rollup(struct seq_file *m, void *v)
 878 {
 879         struct proc_maps_private *priv = m->private;
 880         struct mem_size_stats mss = {};
 881         struct mm_struct *mm = priv->mm;
 882         struct vm_area_struct *vma;
 883         unsigned long vma_start = 0, last_vma_end = 0;
 884         int ret = 0;
 885         VMA_ITERATOR(vmi, mm, 0);
 886
 887         priv->task = get_proc_task(priv->inode);
 888         if (!priv->task)
 889                 return -ESRCH;
 890
 891         if (!mm || !mmget_not_zero(mm)) {
 892                 ret = -ESRCH;
 893                 goto out_put_task;
 894         }
 895
 896         ret = mmap_read_lock_killable(mm);
 897         if (ret)
 898                 goto out_put_mm;
 899
 900         hold_task_mempolicy(priv);
 901         vma = vma_next(&vmi);
 902
 903         if (unlikely(!vma))
 904                 goto empty_set;
 905
 906         vma_start = vma->vm_start;
 907         do {
 908                 smap_gather_stats(vma, &mss, 0);
 909                 last_vma_end = vma->vm_end;
 910
 911                 /*
 912                  * Release mmap_lock temporarily if someone wants to
 913                  * access it for write request.
 914                  */
 915                 if (mmap_lock_is_contended(mm)) {
 916                         vma_iter_invalidate(&vmi);
 917                         mmap_read_unlock(mm);
 918                         ret = mmap_read_lock_killable(mm);
 919                         if (ret) {
 920                                 release_task_mempolicy(priv);
 921                                 goto out_put_mm;
 922                         }
 923
 924                         /*
 925                          * After dropping the lock, there are four cases to
 926                          * consider. See the following example for explanation.
 927                          *
 928                          *   +------+------+-----------+
 929                          *   | VMA1 | VMA2 | VMA3      |
 930                          *   +------+------+-----------+
 931                          *   |      |      |           |
 932                          *  4k     8k     16k         400k
 933                          *
 934                          * Suppose we drop the lock after reading VMA2 due to
 935                          * contention, then we get:
 936                          *
 937                          *      last_vma_end = 16k
 938                          *
 939                          * 1) VMA2 is freed, but VMA3 exists:
 940                          *
 941                          *    vma_next(vmi) will return VMA3.
 942                          *    In this case, just continue from VMA3.
 943                          *
 944                          * 2) VMA2 still exists:
 945                          *
 946                          *    vma_next(vmi) will return VMA3.
 947                          *    In this case, just continue from VMA3.
 948                          *
 949                          * 3) No more VMAs can be found:
 950                          *
 951                          *    vma_next(vmi) will return NULL.
 952                          *    No more things to do, just break.
 953                          *
 954                          * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
 955                          *
 956                          *    vma_next(vmi) will return VMA' whose range
 957                          *    contains last_vma_end.
 958                          *    Iterate VMA' from last_vma_end.
 959                          */
 960                         vma = vma_next(&vmi);
 961                         /* Case 3 above */
 962                         if (!vma)
 963                                 break;
 964
 965                         /* Case 1 and 2 above */
 966                         if (vma->vm_start >= last_vma_end)
 967                                 continue;
 968
 969                         /* Case 4 above */
 970                         if (vma->vm_end > last_vma_end)
 971                                 smap_gather_stats(vma, &mss, last_vma_end);
 972                 }
 973         } for_each_vma(vmi, vma);
 974
 975 empty_set:
 976         show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
 977         seq_pad(m, ' ');
 978         seq_puts(m, "[rollup]\n");
 979
 980         __show_smap(m, &mss, true);
 981
 982         release_task_mempolicy(priv);
 983         mmap_read_unlock(mm);
 984
 985 out_put_mm:
 986         mmput(mm);
 987 out_put_task:
 988         put_task_struct(priv->task);
 989         priv->task = NULL;
 990
 991         return ret;
 992 }
 993 #undef SEQ_PUT_DEC
 994
 995 static const struct seq_operations proc_pid_smaps_op = {
 996         .start  = m_start,
 997         .next   = m_next,
 998         .stop   = m_stop,
 999         .show   = show_smap
1000 };
1001
1002 static int pid_smaps_open(struct inode *inode, struct file *file)
1003 {
1004         return do_maps_open(inode, file, &proc_pid_smaps_op);
1005 }
1006
1007 static int smaps_rollup_open(struct inode *inode, struct file *file)
1008 {
1009         int ret;
1010         struct proc_maps_private *priv;
1011
1012         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1013         if (!priv)
1014                 return -ENOMEM;
1015
1016         ret = single_open(file, show_smaps_rollup, priv);
1017         if (ret)
1018                 goto out_free;
1019
1020         priv->inode = inode;
1021         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1022         if (IS_ERR(priv->mm)) {
1023                 ret = PTR_ERR(priv->mm);
1024
1025                 single_release(inode, file);
1026                 goto out_free;
1027         }
1028
1029         return 0;
1030
1031 out_free:
1032         kfree(priv);
1033         return ret;
1034 }
1035
1036 static int smaps_rollup_release(struct inode *inode, struct file *file)
1037 {
1038         struct seq_file *seq = file->private_data;
1039         struct proc_maps_private *priv = seq->private;
1040
1041         if (priv->mm)
1042                 mmdrop(priv->mm);
1043
1044         kfree(priv);
1045         return single_release(inode, file);
1046 }
1047
1048 const struct file_operations proc_pid_smaps_operations = {
1049         .open           = pid_smaps_open,
1050         .read           = seq_read,
1051         .llseek         = seq_lseek,
1052         .release        = proc_map_release,
1053 };
1054
1055 const struct file_operations proc_pid_smaps_rollup_operations = {
1056         .open           = smaps_rollup_open,
1057         .read           = seq_read,
1058         .llseek         = seq_lseek,
1059         .release        = smaps_rollup_release,
1060 };
1061
1062 enum clear_refs_types {
1063         CLEAR_REFS_ALL = 1,
1064         CLEAR_REFS_ANON,
1065         CLEAR_REFS_MAPPED,
1066         CLEAR_REFS_SOFT_DIRTY,
1067         CLEAR_REFS_MM_HIWATER_RSS,
1068         CLEAR_REFS_LAST,
1069 };
1070
1071 struct clear_refs_private {
1072         enum clear_refs_types type;
1073 };
1074
1075 #ifdef CONFIG_MEM_SOFT_DIRTY
1076
1077 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1078 {
1079         struct page *page;
1080
1081         if (!pte_write(pte))
1082                 return false;
1083         if (!is_cow_mapping(vma->vm_flags))
1084                 return false;
1085         if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1086                 return false;
1087         page = vm_normal_page(vma, addr, pte);
1088         if (!page)
1089                 return false;
1090         return page_maybe_dma_pinned(page);
1091 }
1092
1093 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1094                 unsigned long addr, pte_t *pte)
1095 {
1096         /*
1097          * The soft-dirty tracker uses #PF-s to catch writes
1098          * to pages, so write-protect the pte as well. See the
1099          * Documentation/admin-guide/mm/soft-dirty.rst for full description
1100          * of how soft-dirty works.
1101          */
1102         pte_t ptent = ptep_get(pte);
1103
1104         if (pte_present(ptent)) {
1105                 pte_t old_pte;
1106
1107                 if (pte_is_pinned(vma, addr, ptent))
1108                         return;
1109                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1110                 ptent = pte_wrprotect(old_pte);
1111                 ptent = pte_clear_soft_dirty(ptent);
1112                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1113         } else if (is_swap_pte(ptent)) {
1114                 ptent = pte_swp_clear_soft_dirty(ptent);
1115                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1116         }
1117 }
1118 #else
1119 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1120                 unsigned long addr, pte_t *pte)
1121 {
1122 }
1123 #endif
1124
1125 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1126 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1127                 unsigned long addr, pmd_t *pmdp)
1128 {
1129         pmd_t old, pmd = *pmdp;
1130
1131         if (pmd_present(pmd)) {
1132                 /* See comment in change_huge_pmd() */
1133                 old = pmdp_invalidate(vma, addr, pmdp);
1134                 if (pmd_dirty(old))
1135                         pmd = pmd_mkdirty(pmd);
1136                 if (pmd_young(old))
1137                         pmd = pmd_mkyoung(pmd);
1138
1139                 pmd = pmd_wrprotect(pmd);
1140                 pmd = pmd_clear_soft_dirty(pmd);
1141
1142                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1143         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1144                 pmd = pmd_swp_clear_soft_dirty(pmd);
1145                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1146         }
1147 }
1148 #else
1149 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1150                 unsigned long addr, pmd_t *pmdp)
1151 {
1152 }
1153 #endif
1154
1155 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1156                                 unsigned long end, struct mm_walk *walk)
1157 {
1158         struct clear_refs_private *cp = walk->private;
1159         struct vm_area_struct *vma = walk->vma;
1160         pte_t *pte, ptent;
1161         spinlock_t *ptl;
1162         struct page *page;
1163
1164         ptl = pmd_trans_huge_lock(pmd, vma);
1165         if (ptl) {
1166                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1167                         clear_soft_dirty_pmd(vma, addr, pmd);
1168                         goto out;
1169                 }
1170
1171                 if (!pmd_present(*pmd))
1172                         goto out;
1173
1174                 page = pmd_page(*pmd);
1175
1176                 /* Clear accessed and referenced bits. */
1177                 pmdp_test_and_clear_young(vma, addr, pmd);
1178                 test_and_clear_page_young(page);
1179                 ClearPageReferenced(page);
1180 out:
1181                 spin_unlock(ptl);
1182                 return 0;
1183         }
1184
1185         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1186         if (!pte) {
1187                 walk->action = ACTION_AGAIN;
1188                 return 0;
1189         }
1190         for (; addr != end; pte++, addr += PAGE_SIZE) {
1191                 ptent = ptep_get(pte);
1192
1193                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1194                         clear_soft_dirty(vma, addr, pte);
1195                         continue;
1196                 }
1197
1198                 if (!pte_present(ptent))
1199                         continue;
1200
1201                 page = vm_normal_page(vma, addr, ptent);
1202                 if (!page)
1203                         continue;
1204
1205                 /* Clear accessed and referenced bits. */
1206                 ptep_test_and_clear_young(vma, addr, pte);
1207                 test_and_clear_page_young(page);
1208                 ClearPageReferenced(page);
1209         }
1210         pte_unmap_unlock(pte - 1, ptl);
1211         cond_resched();
1212         return 0;
1213 }
1214
1215 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1216                                 struct mm_walk *walk)
1217 {
1218         struct clear_refs_private *cp = walk->private;
1219         struct vm_area_struct *vma = walk->vma;
1220
1221         if (vma->vm_flags & VM_PFNMAP)
1222                 return 1;
1223
1224         /*
1225          * Writing 1 to /proc/pid/clear_refs affects all pages.
1226          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1227          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1228          * Writing 4 to /proc/pid/clear_refs affects all pages.
1229          */
1230         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1231                 return 1;
1232         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1233                 return 1;
1234         return 0;
1235 }
1236
1237 static const struct mm_walk_ops clear_refs_walk_ops = {
1238         .pmd_entry              = clear_refs_pte_range,
1239         .test_walk              = clear_refs_test_walk,
1240         .walk_lock              = PGWALK_WRLOCK,
1241 };
1242
1243 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1244                                 size_t count, loff_t *ppos)
1245 {
1246         struct task_struct *task;
1247         char buffer[PROC_NUMBUF] = {};
1248         struct mm_struct *mm;
1249         struct vm_area_struct *vma;
1250         enum clear_refs_types type;
1251         int itype;
1252         int rv;
1253
1254         if (count > sizeof(buffer) - 1)
1255                 count = sizeof(buffer) - 1;
1256         if (copy_from_user(buffer, buf, count))
1257                 return -EFAULT;
1258         rv = kstrtoint(strstrip(buffer), 10, &itype);
1259         if (rv < 0)
1260                 return rv;
1261         type = (enum clear_refs_types)itype;
1262         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1263                 return -EINVAL;
1264
1265         task = get_proc_task(file_inode(file));
1266         if (!task)
1267                 return -ESRCH;
1268         mm = get_task_mm(task);
1269         if (mm) {
1270                 VMA_ITERATOR(vmi, mm, 0);
1271                 struct mmu_notifier_range range;
1272                 struct clear_refs_private cp = {
1273                         .type = type,
1274                 };
1275
1276                 if (mmap_write_lock_killable(mm)) {
1277                         count = -EINTR;
1278                         goto out_mm;
1279                 }
1280                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1281                         /*
1282                          * Writing 5 to /proc/pid/clear_refs resets the peak
1283                          * resident set size to this mm's current rss value.
1284                          */
1285                         reset_mm_hiwater_rss(mm);
1286                         goto out_unlock;
1287                 }
1288
1289                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1290                         for_each_vma(vmi, vma) {
1291                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1292                                         continue;
1293                                 vm_flags_clear(vma, VM_SOFTDIRTY);
1294                                 vma_set_page_prot(vma);
1295                         }
1296
1297                         inc_tlb_flush_pending(mm);
1298                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1299                                                 0, mm, 0, -1UL);
1300                         mmu_notifier_invalidate_range_start(&range);
1301                 }
1302                 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1303                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1304                         mmu_notifier_invalidate_range_end(&range);
1305                         flush_tlb_mm(mm);
1306                         dec_tlb_flush_pending(mm);
1307                 }
1308 out_unlock:
1309                 mmap_write_unlock(mm);
1310 out_mm:
1311                 mmput(mm);
1312         }
1313         put_task_struct(task);
1314
1315         return count;
1316 }
1317
1318 const struct file_operations proc_clear_refs_operations = {
1319         .write          = clear_refs_write,
1320         .llseek         = noop_llseek,
1321 };
1322
1323 typedef struct {
1324         u64 pme;
1325 } pagemap_entry_t;
1326
1327 struct pagemapread {
1328         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1329         pagemap_entry_t *buffer;
1330         bool show_pfn;
1331 };
1332
1333 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1334 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1335
1336 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1337 #define PM_PFRAME_BITS          55
1338 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1339 #define PM_SOFT_DIRTY           BIT_ULL(55)
1340 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1341 #define PM_UFFD_WP              BIT_ULL(57)
1342 #define PM_FILE                 BIT_ULL(61)
1343 #define PM_SWAP                 BIT_ULL(62)
1344 #define PM_PRESENT              BIT_ULL(63)
1345
1346 #define PM_END_OF_BUFFER    1
1347
1348 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1349 {
1350         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1351 }
1352
1353 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1354                           struct pagemapread *pm)
1355 {
1356         pm->buffer[pm->pos++] = *pme;
1357         if (pm->pos >= pm->len)
1358                 return PM_END_OF_BUFFER;
1359         return 0;
1360 }
1361
1362 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1363                             __always_unused int depth, struct mm_walk *walk)
1364 {
1365         struct pagemapread *pm = walk->private;
1366         unsigned long addr = start;
1367         int err = 0;
1368
1369         while (addr < end) {
1370                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1371                 pagemap_entry_t pme = make_pme(0, 0);
1372                 /* End of address space hole, which we mark as non-present. */
1373                 unsigned long hole_end;
1374
1375                 if (vma)
1376                         hole_end = min(end, vma->vm_start);
1377                 else
1378                         hole_end = end;
1379
1380                 for (; addr < hole_end; addr += PAGE_SIZE) {
1381                         err = add_to_pagemap(addr, &pme, pm);
1382                         if (err)
1383                                 goto out;
1384                 }
1385
1386                 if (!vma)
1387                         break;
1388
1389                 /* Addresses in the VMA. */
1390                 if (vma->vm_flags & VM_SOFTDIRTY)
1391                         pme = make_pme(0, PM_SOFT_DIRTY);
1392                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1393                         err = add_to_pagemap(addr, &pme, pm);
1394                         if (err)
1395                                 goto out;
1396                 }
1397         }
1398 out:
1399         return err;
1400 }
1401
1402 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1403                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1404 {
1405         u64 frame = 0, flags = 0;
1406         struct page *page = NULL;
1407         bool migration = false;
1408
1409         if (pte_present(pte)) {
1410                 if (pm->show_pfn)
1411                         frame = pte_pfn(pte);
1412                 flags |= PM_PRESENT;
1413                 page = vm_normal_page(vma, addr, pte);
1414                 if (pte_soft_dirty(pte))
1415                         flags |= PM_SOFT_DIRTY;
1416                 if (pte_uffd_wp(pte))
1417                         flags |= PM_UFFD_WP;
1418         } else if (is_swap_pte(pte)) {
1419                 swp_entry_t entry;
1420                 if (pte_swp_soft_dirty(pte))
1421                         flags |= PM_SOFT_DIRTY;
1422                 if (pte_swp_uffd_wp(pte))
1423                         flags |= PM_UFFD_WP;
1424                 entry = pte_to_swp_entry(pte);
1425                 if (pm->show_pfn) {
1426                         pgoff_t offset;
1427                         /*
1428                          * For PFN swap offsets, keeping the offset field
1429                          * to be PFN only to be compatible with old smaps.
1430                          */
1431                         if (is_pfn_swap_entry(entry))
1432                                 offset = swp_offset_pfn(entry);
1433                         else
1434                                 offset = swp_offset(entry);
1435                         frame = swp_type(entry) |
1436                             (offset << MAX_SWAPFILES_SHIFT);
1437                 }
1438                 flags |= PM_SWAP;
1439                 migration = is_migration_entry(entry);
1440                 if (is_pfn_swap_entry(entry))
1441                         page = pfn_swap_entry_to_page(entry);
1442                 if (pte_marker_entry_uffd_wp(entry))
1443                         flags |= PM_UFFD_WP;
1444         }
1445
1446         if (page && !PageAnon(page))
1447                 flags |= PM_FILE;
1448         if (page && !migration && page_mapcount(page) == 1)
1449                 flags |= PM_MMAP_EXCLUSIVE;
1450         if (vma->vm_flags & VM_SOFTDIRTY)
1451                 flags |= PM_SOFT_DIRTY;
1452
1453         return make_pme(frame, flags);
1454 }
1455
1456 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1457                              struct mm_walk *walk)
1458 {
1459         struct vm_area_struct *vma = walk->vma;
1460         struct pagemapread *pm = walk->private;
1461         spinlock_t *ptl;
1462         pte_t *pte, *orig_pte;
1463         int err = 0;
1464 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1465         bool migration = false;
1466
1467         ptl = pmd_trans_huge_lock(pmdp, vma);
1468         if (ptl) {
1469                 u64 flags = 0, frame = 0;
1470                 pmd_t pmd = *pmdp;
1471                 struct page *page = NULL;
1472
1473                 if (vma->vm_flags & VM_SOFTDIRTY)
1474                         flags |= PM_SOFT_DIRTY;
1475
1476                 if (pmd_present(pmd)) {
1477                         page = pmd_page(pmd);
1478
1479                         flags |= PM_PRESENT;
1480                         if (pmd_soft_dirty(pmd))
1481                                 flags |= PM_SOFT_DIRTY;
1482                         if (pmd_uffd_wp(pmd))
1483                                 flags |= PM_UFFD_WP;
1484                         if (pm->show_pfn)
1485                                 frame = pmd_pfn(pmd) +
1486                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1487                 }
1488 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1489                 else if (is_swap_pmd(pmd)) {
1490                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1491                         unsigned long offset;
1492
1493                         if (pm->show_pfn) {
1494                                 if (is_pfn_swap_entry(entry))
1495                                         offset = swp_offset_pfn(entry);
1496                                 else
1497                                         offset = swp_offset(entry);
1498                                 offset = offset +
1499                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1500                                 frame = swp_type(entry) |
1501                                         (offset << MAX_SWAPFILES_SHIFT);
1502                         }
1503                         flags |= PM_SWAP;
1504                         if (pmd_swp_soft_dirty(pmd))
1505                                 flags |= PM_SOFT_DIRTY;
1506                         if (pmd_swp_uffd_wp(pmd))
1507                                 flags |= PM_UFFD_WP;
1508                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1509                         migration = is_migration_entry(entry);
1510                         page = pfn_swap_entry_to_page(entry);
1511                 }
1512 #endif
1513
1514                 if (page && !migration && page_mapcount(page) == 1)
1515                         flags |= PM_MMAP_EXCLUSIVE;
1516
1517                 for (; addr != end; addr += PAGE_SIZE) {
1518                         pagemap_entry_t pme = make_pme(frame, flags);
1519
1520                         err = add_to_pagemap(addr, &pme, pm);
1521                         if (err)
1522                                 break;
1523                         if (pm->show_pfn) {
1524                                 if (flags & PM_PRESENT)
1525                                         frame++;
1526                                 else if (flags & PM_SWAP)
1527                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1528                         }
1529                 }
1530                 spin_unlock(ptl);
1531                 return err;
1532         }
1533 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1534
1535         /*
1536          * We can assume that @vma always points to a valid one and @end never
1537          * goes beyond vma->vm_end.
1538          */
1539         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1540         if (!pte) {
1541                 walk->action = ACTION_AGAIN;
1542                 return err;
1543         }
1544         for (; addr < end; pte++, addr += PAGE_SIZE) {
1545                 pagemap_entry_t pme;
1546
1547                 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte));
1548                 err = add_to_pagemap(addr, &pme, pm);
1549                 if (err)
1550                         break;
1551         }
1552         pte_unmap_unlock(orig_pte, ptl);
1553
1554         cond_resched();
1555
1556         return err;
1557 }
1558
1559 #ifdef CONFIG_HUGETLB_PAGE
1560 /* This function walks within one hugetlb entry in the single call */
1561 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1562                                  unsigned long addr, unsigned long end,
1563                                  struct mm_walk *walk)
1564 {
1565         struct pagemapread *pm = walk->private;
1566         struct vm_area_struct *vma = walk->vma;
1567         u64 flags = 0, frame = 0;
1568         int err = 0;
1569         pte_t pte;
1570
1571         if (vma->vm_flags & VM_SOFTDIRTY)
1572                 flags |= PM_SOFT_DIRTY;
1573
1574         pte = huge_ptep_get(ptep);
1575         if (pte_present(pte)) {
1576                 struct page *page = pte_page(pte);
1577
1578                 if (!PageAnon(page))
1579                         flags |= PM_FILE;
1580
1581                 if (page_mapcount(page) == 1)
1582                         flags |= PM_MMAP_EXCLUSIVE;
1583
1584                 if (huge_pte_uffd_wp(pte))
1585                         flags |= PM_UFFD_WP;
1586
1587                 flags |= PM_PRESENT;
1588                 if (pm->show_pfn)
1589                         frame = pte_pfn(pte) +
1590                                 ((addr & ~hmask) >> PAGE_SHIFT);
1591         } else if (pte_swp_uffd_wp_any(pte)) {
1592                 flags |= PM_UFFD_WP;
1593         }
1594
1595         for (; addr != end; addr += PAGE_SIZE) {
1596                 pagemap_entry_t pme = make_pme(frame, flags);
1597
1598                 err = add_to_pagemap(addr, &pme, pm);
1599                 if (err)
1600                         return err;
1601                 if (pm->show_pfn && (flags & PM_PRESENT))
1602                         frame++;
1603         }
1604
1605         cond_resched();
1606
1607         return err;
1608 }
1609 #else
1610 #define pagemap_hugetlb_range   NULL
1611 #endif /* HUGETLB_PAGE */
1612
1613 static const struct mm_walk_ops pagemap_ops = {
1614         .pmd_entry      = pagemap_pmd_range,
1615         .pte_hole       = pagemap_pte_hole,
1616         .hugetlb_entry  = pagemap_hugetlb_range,
1617         .walk_lock      = PGWALK_RDLOCK,
1618 };
1619
1620 /*
1621  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1622  *
1623  * For each page in the address space, this file contains one 64-bit entry
1624  * consisting of the following:
1625  *
1626  * Bits 0-54  page frame number (PFN) if present
1627  * Bits 0-4   swap type if swapped
1628  * Bits 5-54  swap offset if swapped
1629  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1630  * Bit  56    page exclusively mapped
1631  * Bit  57    pte is uffd-wp write-protected
1632  * Bits 58-60 zero
1633  * Bit  61    page is file-page or shared-anon
1634  * Bit  62    page swapped
1635  * Bit  63    page present
1636  *
1637  * If the page is not present but in swap, then the PFN contains an
1638  * encoding of the swap file number and the page's offset into the
1639  * swap. Unmapped pages return a null PFN. This allows determining
1640  * precisely which pages are mapped (or in swap) and comparing mapped
1641  * pages between processes.
1642  *
1643  * Efficient users of this interface will use /proc/pid/maps to
1644  * determine which areas of memory are actually mapped and llseek to
1645  * skip over unmapped regions.
1646  */
1647 static ssize_t pagemap_read(struct file *file, char __user *buf,
1648                             size_t count, loff_t *ppos)
1649 {
1650         struct mm_struct *mm = file->private_data;
1651         struct pagemapread pm;
1652         unsigned long src;
1653         unsigned long svpfn;
1654         unsigned long start_vaddr;
1655         unsigned long end_vaddr;
1656         int ret = 0, copied = 0;
1657
1658         if (!mm || !mmget_not_zero(mm))
1659                 goto out;
1660
1661         ret = -EINVAL;
1662         /* file position must be aligned */
1663         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1664                 goto out_mm;
1665
1666         ret = 0;
1667         if (!count)
1668                 goto out_mm;
1669
1670         /* do not disclose physical addresses: attack vector */
1671         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1672
1673         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1674         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1675         ret = -ENOMEM;
1676         if (!pm.buffer)
1677                 goto out_mm;
1678
1679         src = *ppos;
1680         svpfn = src / PM_ENTRY_BYTES;
1681         end_vaddr = mm->task_size;
1682
1683         /* watch out for wraparound */
1684         start_vaddr = end_vaddr;
1685         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
1686                 unsigned long end;
1687
1688                 ret = mmap_read_lock_killable(mm);
1689                 if (ret)
1690                         goto out_free;
1691                 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
1692                 mmap_read_unlock(mm);
1693
1694                 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
1695                 if (end >= start_vaddr && end < mm->task_size)
1696                         end_vaddr = end;
1697         }
1698
1699         /* Ensure the address is inside the task */
1700         if (start_vaddr > mm->task_size)
1701                 start_vaddr = end_vaddr;
1702
1703         ret = 0;
1704         while (count && (start_vaddr < end_vaddr)) {
1705                 int len;
1706                 unsigned long end;
1707
1708                 pm.pos = 0;
1709                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1710                 /* overflow ? */
1711                 if (end < start_vaddr || end > end_vaddr)
1712                         end = end_vaddr;
1713                 ret = mmap_read_lock_killable(mm);
1714                 if (ret)
1715                         goto out_free;
1716                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1717                 mmap_read_unlock(mm);
1718                 start_vaddr = end;
1719
1720                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1721                 if (copy_to_user(buf, pm.buffer, len)) {
1722                         ret = -EFAULT;
1723                         goto out_free;
1724                 }
1725                 copied += len;
1726                 buf += len;
1727                 count -= len;
1728         }
1729         *ppos += copied;
1730         if (!ret || ret == PM_END_OF_BUFFER)
1731                 ret = copied;
1732
1733 out_free:
1734         kfree(pm.buffer);
1735 out_mm:
1736         mmput(mm);
1737 out:
1738         return ret;
1739 }
1740
1741 static int pagemap_open(struct inode *inode, struct file *file)
1742 {
1743         struct mm_struct *mm;
1744
1745         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1746         if (IS_ERR(mm))
1747                 return PTR_ERR(mm);
1748         file->private_data = mm;
1749         return 0;
1750 }
1751
1752 static int pagemap_release(struct inode *inode, struct file *file)
1753 {
1754         struct mm_struct *mm = file->private_data;
1755
1756         if (mm)
1757                 mmdrop(mm);
1758         return 0;
1759 }
1760
1761 #define PM_SCAN_CATEGORIES      (PAGE_IS_WPALLOWED | PAGE_IS_WRITTEN |  \
1762                                  PAGE_IS_FILE | PAGE_IS_PRESENT |       \
1763                                  PAGE_IS_SWAPPED | PAGE_IS_PFNZERO |    \
1764                                  PAGE_IS_HUGE)
1765 #define PM_SCAN_FLAGS           (PM_SCAN_WP_MATCHING | PM_SCAN_CHECK_WPASYNC)
1766
1767 struct pagemap_scan_private {
1768         struct pm_scan_arg arg;
1769         unsigned long masks_of_interest, cur_vma_category;
1770         struct page_region *vec_buf;
1771         unsigned long vec_buf_len, vec_buf_index, found_pages;
1772         struct page_region __user *vec_out;
1773 };
1774
1775 static unsigned long pagemap_page_category(struct pagemap_scan_private *p,
1776                                            struct vm_area_struct *vma,
1777                                            unsigned long addr, pte_t pte)
1778 {
1779         unsigned long categories = 0;
1780
1781         if (pte_present(pte)) {
1782                 struct page *page;
1783
1784                 categories |= PAGE_IS_PRESENT;
1785                 if (!pte_uffd_wp(pte))
1786                         categories |= PAGE_IS_WRITTEN;
1787
1788                 if (p->masks_of_interest & PAGE_IS_FILE) {
1789                         page = vm_normal_page(vma, addr, pte);
1790                         if (page && !PageAnon(page))
1791                                 categories |= PAGE_IS_FILE;
1792                 }
1793
1794                 if (is_zero_pfn(pte_pfn(pte)))
1795                         categories |= PAGE_IS_PFNZERO;
1796         } else if (is_swap_pte(pte)) {
1797                 swp_entry_t swp;
1798
1799                 categories |= PAGE_IS_SWAPPED;
1800                 if (!pte_swp_uffd_wp_any(pte))
1801                         categories |= PAGE_IS_WRITTEN;
1802
1803                 if (p->masks_of_interest & PAGE_IS_FILE) {
1804                         swp = pte_to_swp_entry(pte);
1805                         if (is_pfn_swap_entry(swp) &&
1806                             !PageAnon(pfn_swap_entry_to_page(swp)))
1807                                 categories |= PAGE_IS_FILE;
1808                 }
1809         }
1810
1811         return categories;
1812 }
1813
1814 static void make_uffd_wp_pte(struct vm_area_struct *vma,
1815                              unsigned long addr, pte_t *pte)
1816 {
1817         pte_t ptent = ptep_get(pte);
1818
1819         if (pte_present(ptent)) {
1820                 pte_t old_pte;
1821
1822                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1823                 ptent = pte_mkuffd_wp(ptent);
1824                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1825         } else if (is_swap_pte(ptent)) {
1826                 ptent = pte_swp_mkuffd_wp(ptent);
1827                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1828         } else {
1829                 set_pte_at(vma->vm_mm, addr, pte,
1830                            make_pte_marker(PTE_MARKER_UFFD_WP));
1831         }
1832 }
1833
1834 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1835 static unsigned long pagemap_thp_category(struct pagemap_scan_private *p,
1836                                           struct vm_area_struct *vma,
1837                                           unsigned long addr, pmd_t pmd)
1838 {
1839         unsigned long categories = PAGE_IS_HUGE;
1840
1841         if (pmd_present(pmd)) {
1842                 struct page *page;
1843
1844                 categories |= PAGE_IS_PRESENT;
1845                 if (!pmd_uffd_wp(pmd))
1846                         categories |= PAGE_IS_WRITTEN;
1847
1848                 if (p->masks_of_interest & PAGE_IS_FILE) {
1849                         page = vm_normal_page_pmd(vma, addr, pmd);
1850                         if (page && !PageAnon(page))
1851                                 categories |= PAGE_IS_FILE;
1852                 }
1853
1854                 if (is_zero_pfn(pmd_pfn(pmd)))
1855                         categories |= PAGE_IS_PFNZERO;
1856         } else if (is_swap_pmd(pmd)) {
1857                 swp_entry_t swp;
1858
1859                 categories |= PAGE_IS_SWAPPED;
1860                 if (!pmd_swp_uffd_wp(pmd))
1861                         categories |= PAGE_IS_WRITTEN;
1862
1863                 if (p->masks_of_interest & PAGE_IS_FILE) {
1864                         swp = pmd_to_swp_entry(pmd);
1865                         if (is_pfn_swap_entry(swp) &&
1866                             !PageAnon(pfn_swap_entry_to_page(swp)))
1867                                 categories |= PAGE_IS_FILE;
1868                 }
1869         }
1870
1871         return categories;
1872 }
1873
1874 static void make_uffd_wp_pmd(struct vm_area_struct *vma,
1875                              unsigned long addr, pmd_t *pmdp)
1876 {
1877         pmd_t old, pmd = *pmdp;
1878
1879         if (pmd_present(pmd)) {
1880                 old = pmdp_invalidate_ad(vma, addr, pmdp);
1881                 pmd = pmd_mkuffd_wp(old);
1882                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1883         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1884                 pmd = pmd_swp_mkuffd_wp(pmd);
1885                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1886         }
1887 }
1888 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1889
1890 #ifdef CONFIG_HUGETLB_PAGE
1891 static unsigned long pagemap_hugetlb_category(pte_t pte)
1892 {
1893         unsigned long categories = PAGE_IS_HUGE;
1894
1895         /*
1896          * According to pagemap_hugetlb_range(), file-backed HugeTLB
1897          * page cannot be swapped. So PAGE_IS_FILE is not checked for
1898          * swapped pages.
1899          */
1900         if (pte_present(pte)) {
1901                 categories |= PAGE_IS_PRESENT;
1902                 if (!huge_pte_uffd_wp(pte))
1903                         categories |= PAGE_IS_WRITTEN;
1904                 if (!PageAnon(pte_page(pte)))
1905                         categories |= PAGE_IS_FILE;
1906                 if (is_zero_pfn(pte_pfn(pte)))
1907                         categories |= PAGE_IS_PFNZERO;
1908         } else if (is_swap_pte(pte)) {
1909                 categories |= PAGE_IS_SWAPPED;
1910                 if (!pte_swp_uffd_wp_any(pte))
1911                         categories |= PAGE_IS_WRITTEN;
1912         }
1913
1914         return categories;
1915 }
1916
1917 static void make_uffd_wp_huge_pte(struct vm_area_struct *vma,
1918                                   unsigned long addr, pte_t *ptep,
1919                                   pte_t ptent)
1920 {
1921         unsigned long psize;
1922
1923         if (is_hugetlb_entry_hwpoisoned(ptent) || is_pte_marker(ptent))
1924                 return;
1925
1926         psize = huge_page_size(hstate_vma(vma));
1927
1928         if (is_hugetlb_entry_migration(ptent))
1929                 set_huge_pte_at(vma->vm_mm, addr, ptep,
1930                                 pte_swp_mkuffd_wp(ptent), psize);
1931         else if (!huge_pte_none(ptent))
1932                 huge_ptep_modify_prot_commit(vma, addr, ptep, ptent,
1933                                              huge_pte_mkuffd_wp(ptent));
1934         else
1935                 set_huge_pte_at(vma->vm_mm, addr, ptep,
1936                                 make_pte_marker(PTE_MARKER_UFFD_WP), psize);
1937 }
1938 #endif /* CONFIG_HUGETLB_PAGE */
1939
1940 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE)
1941 static void pagemap_scan_backout_range(struct pagemap_scan_private *p,
1942                                        unsigned long addr, unsigned long end)
1943 {
1944         struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
1945
1946         if (cur_buf->start != addr)
1947                 cur_buf->end = addr;
1948         else
1949                 cur_buf->start = cur_buf->end = 0;
1950
1951         p->found_pages -= (end - addr) / PAGE_SIZE;
1952 }
1953 #endif
1954
1955 static bool pagemap_scan_is_interesting_page(unsigned long categories,
1956                                              const struct pagemap_scan_private *p)
1957 {
1958         categories ^= p->arg.category_inverted;
1959         if ((categories & p->arg.category_mask) != p->arg.category_mask)
1960                 return false;
1961         if (p->arg.category_anyof_mask && !(categories & p->arg.category_anyof_mask))
1962                 return false;
1963
1964         return true;
1965 }
1966
1967 static bool pagemap_scan_is_interesting_vma(unsigned long categories,
1968                                             const struct pagemap_scan_private *p)
1969 {
1970         unsigned long required = p->arg.category_mask & PAGE_IS_WPALLOWED;
1971
1972         categories ^= p->arg.category_inverted;
1973         if ((categories & required) != required)
1974                 return false;
1975
1976         return true;
1977 }
1978
1979 static int pagemap_scan_test_walk(unsigned long start, unsigned long end,
1980                                   struct mm_walk *walk)
1981 {
1982         struct pagemap_scan_private *p = walk->private;
1983         struct vm_area_struct *vma = walk->vma;
1984         unsigned long vma_category = 0;
1985         bool wp_allowed = userfaultfd_wp_async(vma) &&
1986             userfaultfd_wp_use_markers(vma);
1987
1988         if (!wp_allowed) {
1989                 /* User requested explicit failure over wp-async capability */
1990                 if (p->arg.flags & PM_SCAN_CHECK_WPASYNC)
1991                         return -EPERM;
1992                 /*
1993                  * User requires wr-protect, and allows silently skipping
1994                  * unsupported vmas.
1995                  */
1996                 if (p->arg.flags & PM_SCAN_WP_MATCHING)
1997                         return 1;
1998                 /*
1999                  * Then the request doesn't involve wr-protects at all,
2000                  * fall through to the rest checks, and allow vma walk.
2001                  */
2002         }
2003
2004         if (vma->vm_flags & VM_PFNMAP)
2005                 return 1;
2006
2007         if (wp_allowed)
2008                 vma_category |= PAGE_IS_WPALLOWED;
2009
2010         if (!pagemap_scan_is_interesting_vma(vma_category, p))
2011                 return 1;
2012
2013         p->cur_vma_category = vma_category;
2014
2015         return 0;
2016 }
2017
2018 static bool pagemap_scan_push_range(unsigned long categories,
2019                                     struct pagemap_scan_private *p,
2020                                     unsigned long addr, unsigned long end)
2021 {
2022         struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
2023
2024         /*
2025          * When there is no output buffer provided at all, the sentinel values
2026          * won't match here. There is no other way for `cur_buf->end` to be
2027          * non-zero other than it being non-empty.
2028          */
2029         if (addr == cur_buf->end && categories == cur_buf->categories) {
2030                 cur_buf->end = end;
2031                 return true;
2032         }
2033
2034         if (cur_buf->end) {
2035                 if (p->vec_buf_index >= p->vec_buf_len - 1)
2036                         return false;
2037
2038                 cur_buf = &p->vec_buf[++p->vec_buf_index];
2039         }
2040
2041         cur_buf->start = addr;
2042         cur_buf->end = end;
2043         cur_buf->categories = categories;
2044
2045         return true;
2046 }
2047
2048 static int pagemap_scan_output(unsigned long categories,
2049                                struct pagemap_scan_private *p,
2050                                unsigned long addr, unsigned long *end)
2051 {
2052         unsigned long n_pages, total_pages;
2053         int ret = 0;
2054
2055         if (!p->vec_buf)
2056                 return 0;
2057
2058         categories &= p->arg.return_mask;
2059
2060         n_pages = (*end - addr) / PAGE_SIZE;
2061         if (check_add_overflow(p->found_pages, n_pages, &total_pages) ||
2062             total_pages > p->arg.max_pages) {
2063                 size_t n_too_much = total_pages - p->arg.max_pages;
2064                 *end -= n_too_much * PAGE_SIZE;
2065                 n_pages -= n_too_much;
2066                 ret = -ENOSPC;
2067         }
2068
2069         if (!pagemap_scan_push_range(categories, p, addr, *end)) {
2070                 *end = addr;
2071                 n_pages = 0;
2072                 ret = -ENOSPC;
2073         }
2074
2075         p->found_pages += n_pages;
2076         if (ret)
2077                 p->arg.walk_end = *end;
2078
2079         return ret;
2080 }
2081
2082 static int pagemap_scan_thp_entry(pmd_t *pmd, unsigned long start,
2083                                   unsigned long end, struct mm_walk *walk)
2084 {
2085 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2086         struct pagemap_scan_private *p = walk->private;
2087         struct vm_area_struct *vma = walk->vma;
2088         unsigned long categories;
2089         spinlock_t *ptl;
2090         int ret = 0;
2091
2092         ptl = pmd_trans_huge_lock(pmd, vma);
2093         if (!ptl)
2094                 return -ENOENT;
2095
2096         categories = p->cur_vma_category |
2097                      pagemap_thp_category(p, vma, start, *pmd);
2098
2099         if (!pagemap_scan_is_interesting_page(categories, p))
2100                 goto out_unlock;
2101
2102         ret = pagemap_scan_output(categories, p, start, &end);
2103         if (start == end)
2104                 goto out_unlock;
2105
2106         if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2107                 goto out_unlock;
2108         if (~categories & PAGE_IS_WRITTEN)
2109                 goto out_unlock;
2110
2111         /*
2112          * Break huge page into small pages if the WP operation
2113          * needs to be performed on a portion of the huge page.
2114          */
2115         if (end != start + HPAGE_SIZE) {
2116                 spin_unlock(ptl);
2117                 split_huge_pmd(vma, pmd, start);
2118                 pagemap_scan_backout_range(p, start, end);
2119                 /* Report as if there was no THP */
2120                 return -ENOENT;
2121         }
2122
2123         make_uffd_wp_pmd(vma, start, pmd);
2124         flush_tlb_range(vma, start, end);
2125 out_unlock:
2126         spin_unlock(ptl);
2127         return ret;
2128 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
2129         return -ENOENT;
2130 #endif
2131 }
2132
2133 static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start,
2134                                   unsigned long end, struct mm_walk *walk)
2135 {
2136         struct pagemap_scan_private *p = walk->private;
2137         struct vm_area_struct *vma = walk->vma;
2138         unsigned long addr, flush_end = 0;
2139         pte_t *pte, *start_pte;
2140         spinlock_t *ptl;
2141         int ret;
2142
2143         arch_enter_lazy_mmu_mode();
2144
2145         ret = pagemap_scan_thp_entry(pmd, start, end, walk);
2146         if (ret != -ENOENT) {
2147                 arch_leave_lazy_mmu_mode();
2148                 return ret;
2149         }
2150
2151         ret = 0;
2152         start_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl);
2153         if (!pte) {
2154                 arch_leave_lazy_mmu_mode();
2155                 walk->action = ACTION_AGAIN;
2156                 return 0;
2157         }
2158
2159         if ((p->arg.flags & PM_SCAN_WP_MATCHING) && !p->vec_out) {
2160                 /* Fast path for performing exclusive WP */
2161                 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2162                         if (pte_uffd_wp(ptep_get(pte)))
2163                                 continue;
2164                         make_uffd_wp_pte(vma, addr, pte);
2165                         if (!flush_end)
2166                                 start = addr;
2167                         flush_end = addr + PAGE_SIZE;
2168                 }
2169                 goto flush_and_return;
2170         }
2171
2172         if (!p->arg.category_anyof_mask && !p->arg.category_inverted &&
2173             p->arg.category_mask == PAGE_IS_WRITTEN &&
2174             p->arg.return_mask == PAGE_IS_WRITTEN) {
2175                 for (addr = start; addr < end; pte++, addr += PAGE_SIZE) {
2176                         unsigned long next = addr + PAGE_SIZE;
2177
2178                         if (pte_uffd_wp(ptep_get(pte)))
2179                                 continue;
2180                         ret = pagemap_scan_output(p->cur_vma_category | PAGE_IS_WRITTEN,
2181                                                   p, addr, &next);
2182                         if (next == addr)
2183                                 break;
2184                         if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2185                                 continue;
2186                         make_uffd_wp_pte(vma, addr, pte);
2187                         if (!flush_end)
2188                                 start = addr;
2189                         flush_end = next;
2190                 }
2191                 goto flush_and_return;
2192         }
2193
2194         for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2195                 unsigned long categories = p->cur_vma_category |
2196                                            pagemap_page_category(p, vma, addr, ptep_get(pte));
2197                 unsigned long next = addr + PAGE_SIZE;
2198
2199                 if (!pagemap_scan_is_interesting_page(categories, p))
2200                         continue;
2201
2202                 ret = pagemap_scan_output(categories, p, addr, &next);
2203                 if (next == addr)
2204                         break;
2205
2206                 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2207                         continue;
2208                 if (~categories & PAGE_IS_WRITTEN)
2209                         continue;
2210
2211                 make_uffd_wp_pte(vma, addr, pte);
2212                 if (!flush_end)
2213                         start = addr;
2214                 flush_end = next;
2215         }
2216
2217 flush_and_return:
2218         if (flush_end)
2219                 flush_tlb_range(vma, start, addr);
2220
2221         pte_unmap_unlock(start_pte, ptl);
2222         arch_leave_lazy_mmu_mode();
2223
2224         cond_resched();
2225         return ret;
2226 }
2227
2228 #ifdef CONFIG_HUGETLB_PAGE
2229 static int pagemap_scan_hugetlb_entry(pte_t *ptep, unsigned long hmask,
2230                                       unsigned long start, unsigned long end,
2231                                       struct mm_walk *walk)
2232 {
2233         struct pagemap_scan_private *p = walk->private;
2234         struct vm_area_struct *vma = walk->vma;
2235         unsigned long categories;
2236         spinlock_t *ptl;
2237         int ret = 0;
2238         pte_t pte;
2239
2240         if (~p->arg.flags & PM_SCAN_WP_MATCHING) {
2241                 /* Go the short route when not write-protecting pages. */
2242
2243                 pte = huge_ptep_get(ptep);
2244                 categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2245
2246                 if (!pagemap_scan_is_interesting_page(categories, p))
2247                         return 0;
2248
2249                 return pagemap_scan_output(categories, p, start, &end);
2250         }
2251
2252         i_mmap_lock_write(vma->vm_file->f_mapping);
2253         ptl = huge_pte_lock(hstate_vma(vma), vma->vm_mm, ptep);
2254
2255         pte = huge_ptep_get(ptep);
2256         categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2257
2258         if (!pagemap_scan_is_interesting_page(categories, p))
2259                 goto out_unlock;
2260
2261         ret = pagemap_scan_output(categories, p, start, &end);
2262         if (start == end)
2263                 goto out_unlock;
2264
2265         if (~categories & PAGE_IS_WRITTEN)
2266                 goto out_unlock;
2267
2268         if (end != start + HPAGE_SIZE) {
2269                 /* Partial HugeTLB page WP isn't possible. */
2270                 pagemap_scan_backout_range(p, start, end);
2271                 p->arg.walk_end = start;
2272                 ret = 0;
2273                 goto out_unlock;
2274         }
2275
2276         make_uffd_wp_huge_pte(vma, start, ptep, pte);
2277         flush_hugetlb_tlb_range(vma, start, end);
2278
2279 out_unlock:
2280         spin_unlock(ptl);
2281         i_mmap_unlock_write(vma->vm_file->f_mapping);
2282
2283         return ret;
2284 }
2285 #else
2286 #define pagemap_scan_hugetlb_entry NULL
2287 #endif
2288
2289 static int pagemap_scan_pte_hole(unsigned long addr, unsigned long end,
2290                                  int depth, struct mm_walk *walk)
2291 {
2292         struct pagemap_scan_private *p = walk->private;
2293         struct vm_area_struct *vma = walk->vma;
2294         int ret, err;
2295
2296         if (!vma || !pagemap_scan_is_interesting_page(p->cur_vma_category, p))
2297                 return 0;
2298
2299         ret = pagemap_scan_output(p->cur_vma_category, p, addr, &end);
2300         if (addr == end)
2301                 return ret;
2302
2303         if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2304                 return ret;
2305
2306         err = uffd_wp_range(vma, addr, end - addr, true);
2307         if (err < 0)
2308                 ret = err;
2309
2310         return ret;
2311 }
2312
2313 static const struct mm_walk_ops pagemap_scan_ops = {
2314         .test_walk = pagemap_scan_test_walk,
2315         .pmd_entry = pagemap_scan_pmd_entry,
2316         .pte_hole = pagemap_scan_pte_hole,
2317         .hugetlb_entry = pagemap_scan_hugetlb_entry,
2318 };
2319
2320 static int pagemap_scan_get_args(struct pm_scan_arg *arg,
2321                                  unsigned long uarg)
2322 {
2323         if (copy_from_user(arg, (void __user *)uarg, sizeof(*arg)))
2324                 return -EFAULT;
2325
2326         if (arg->size != sizeof(struct pm_scan_arg))
2327                 return -EINVAL;
2328
2329         /* Validate requested features */
2330         if (arg->flags & ~PM_SCAN_FLAGS)
2331                 return -EINVAL;
2332         if ((arg->category_inverted | arg->category_mask |
2333              arg->category_anyof_mask | arg->return_mask) & ~PM_SCAN_CATEGORIES)
2334                 return -EINVAL;
2335
2336         arg->start = untagged_addr((unsigned long)arg->start);
2337         arg->end = untagged_addr((unsigned long)arg->end);
2338         arg->vec = untagged_addr((unsigned long)arg->vec);
2339
2340         /* Validate memory pointers */
2341         if (!IS_ALIGNED(arg->start, PAGE_SIZE))
2342                 return -EINVAL;
2343         if (!access_ok((void __user *)(long)arg->start, arg->end - arg->start))
2344                 return -EFAULT;
2345         if (!arg->vec && arg->vec_len)
2346                 return -EINVAL;
2347         if (arg->vec && !access_ok((void __user *)(long)arg->vec,
2348                               arg->vec_len * sizeof(struct page_region)))
2349                 return -EFAULT;
2350
2351         /* Fixup default values */
2352         arg->end = ALIGN(arg->end, PAGE_SIZE);
2353         arg->walk_end = 0;
2354         if (!arg->max_pages)
2355                 arg->max_pages = ULONG_MAX;
2356
2357         return 0;
2358 }
2359
2360 static int pagemap_scan_writeback_args(struct pm_scan_arg *arg,
2361                                        unsigned long uargl)
2362 {
2363         struct pm_scan_arg __user *uarg = (void __user *)uargl;
2364
2365         if (copy_to_user(&uarg->walk_end, &arg->walk_end, sizeof(arg->walk_end)))
2366                 return -EFAULT;
2367
2368         return 0;
2369 }
2370
2371 static int pagemap_scan_init_bounce_buffer(struct pagemap_scan_private *p)
2372 {
2373         if (!p->arg.vec_len)
2374                 return 0;
2375
2376         p->vec_buf_len = min_t(size_t, PAGEMAP_WALK_SIZE >> PAGE_SHIFT,
2377                                p->arg.vec_len);
2378         p->vec_buf = kmalloc_array(p->vec_buf_len, sizeof(*p->vec_buf),
2379                                    GFP_KERNEL);
2380         if (!p->vec_buf)
2381                 return -ENOMEM;
2382
2383         p->vec_buf->start = p->vec_buf->end = 0;
2384         p->vec_out = (struct page_region __user *)(long)p->arg.vec;
2385
2386         return 0;
2387 }
2388
2389 static long pagemap_scan_flush_buffer(struct pagemap_scan_private *p)
2390 {
2391         const struct page_region *buf = p->vec_buf;
2392         long n = p->vec_buf_index;
2393
2394         if (!p->vec_buf)
2395                 return 0;
2396
2397         if (buf[n].end != buf[n].start)
2398                 n++;
2399
2400         if (!n)
2401                 return 0;
2402
2403         if (copy_to_user(p->vec_out, buf, n * sizeof(*buf)))
2404                 return -EFAULT;
2405
2406         p->arg.vec_len -= n;
2407         p->vec_out += n;
2408
2409         p->vec_buf_index = 0;
2410         p->vec_buf_len = min_t(size_t, p->vec_buf_len, p->arg.vec_len);
2411         p->vec_buf->start = p->vec_buf->end = 0;
2412
2413         return n;
2414 }
2415
2416 static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg)
2417 {
2418         struct mmu_notifier_range range;
2419         struct pagemap_scan_private p = {0};
2420         unsigned long walk_start;
2421         size_t n_ranges_out = 0;
2422         int ret;
2423
2424         ret = pagemap_scan_get_args(&p.arg, uarg);
2425         if (ret)
2426                 return ret;
2427
2428         p.masks_of_interest = p.arg.category_mask | p.arg.category_anyof_mask |
2429                               p.arg.return_mask;
2430         ret = pagemap_scan_init_bounce_buffer(&p);
2431         if (ret)
2432                 return ret;
2433
2434         /* Protection change for the range is going to happen. */
2435         if (p.arg.flags & PM_SCAN_WP_MATCHING) {
2436                 mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_VMA, 0,
2437                                         mm, p.arg.start, p.arg.end);
2438                 mmu_notifier_invalidate_range_start(&range);
2439         }
2440
2441         for (walk_start = p.arg.start; walk_start < p.arg.end;
2442                         walk_start = p.arg.walk_end) {
2443                 long n_out;
2444
2445                 if (fatal_signal_pending(current)) {
2446                         ret = -EINTR;
2447                         break;
2448                 }
2449
2450                 ret = mmap_read_lock_killable(mm);
2451                 if (ret)
2452                         break;
2453                 ret = walk_page_range(mm, walk_start, p.arg.end,
2454                                       &pagemap_scan_ops, &p);
2455                 mmap_read_unlock(mm);
2456
2457                 n_out = pagemap_scan_flush_buffer(&p);
2458                 if (n_out < 0)
2459                         ret = n_out;
2460                 else
2461                         n_ranges_out += n_out;
2462
2463                 if (ret != -ENOSPC)
2464                         break;
2465
2466                 if (p.arg.vec_len == 0 || p.found_pages == p.arg.max_pages)
2467                         break;
2468         }
2469
2470         /* ENOSPC signifies early stop (buffer full) from the walk. */
2471         if (!ret || ret == -ENOSPC)
2472                 ret = n_ranges_out;
2473
2474         /* The walk_end isn't set when ret is zero */
2475         if (!p.arg.walk_end)
2476                 p.arg.walk_end = p.arg.end;
2477         if (pagemap_scan_writeback_args(&p.arg, uarg))
2478                 ret = -EFAULT;
2479
2480         if (p.arg.flags & PM_SCAN_WP_MATCHING)
2481                 mmu_notifier_invalidate_range_end(&range);
2482
2483         kfree(p.vec_buf);
2484         return ret;
2485 }
2486
2487 static long do_pagemap_cmd(struct file *file, unsigned int cmd,
2488                            unsigned long arg)
2489 {
2490         struct mm_struct *mm = file->private_data;
2491
2492         switch (cmd) {
2493         case PAGEMAP_SCAN:
2494                 return do_pagemap_scan(mm, arg);
2495
2496         default:
2497                 return -EINVAL;
2498         }
2499 }
2500
2501 const struct file_operations proc_pagemap_operations = {
2502         .llseek         = mem_lseek, /* borrow this */
2503         .read           = pagemap_read,
2504         .open           = pagemap_open,
2505         .release        = pagemap_release,
2506         .unlocked_ioctl = do_pagemap_cmd,
2507         .compat_ioctl   = do_pagemap_cmd,
2508 };
2509 #endif /* CONFIG_PROC_PAGE_MONITOR */
2510
2511 #ifdef CONFIG_NUMA
2512
2513 struct numa_maps {
2514         unsigned long pages;
2515         unsigned long anon;
2516         unsigned long active;
2517         unsigned long writeback;
2518         unsigned long mapcount_max;
2519         unsigned long dirty;
2520         unsigned long swapcache;
2521         unsigned long node[MAX_NUMNODES];
2522 };
2523
2524 struct numa_maps_private {
2525         struct proc_maps_private proc_maps;
2526         struct numa_maps md;
2527 };
2528
2529 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
2530                         unsigned long nr_pages)
2531 {
2532         int count = page_mapcount(page);
2533
2534         md->pages += nr_pages;
2535         if (pte_dirty || PageDirty(page))
2536                 md->dirty += nr_pages;
2537
2538         if (PageSwapCache(page))
2539                 md->swapcache += nr_pages;
2540
2541         if (PageActive(page) || PageUnevictable(page))
2542                 md->active += nr_pages;
2543
2544         if (PageWriteback(page))
2545                 md->writeback += nr_pages;
2546
2547         if (PageAnon(page))
2548                 md->anon += nr_pages;
2549
2550         if (count > md->mapcount_max)
2551                 md->mapcount_max = count;
2552
2553         md->node[page_to_nid(page)] += nr_pages;
2554 }
2555
2556 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
2557                 unsigned long addr)
2558 {
2559         struct page *page;
2560         int nid;
2561
2562         if (!pte_present(pte))
2563                 return NULL;
2564
2565         page = vm_normal_page(vma, addr, pte);
2566         if (!page || is_zone_device_page(page))
2567                 return NULL;
2568
2569         if (PageReserved(page))
2570                 return NULL;
2571
2572         nid = page_to_nid(page);
2573         if (!node_isset(nid, node_states[N_MEMORY]))
2574                 return NULL;
2575
2576         return page;
2577 }
2578
2579 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2580 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
2581                                               struct vm_area_struct *vma,
2582                                               unsigned long addr)
2583 {
2584         struct page *page;
2585         int nid;
2586
2587         if (!pmd_present(pmd))
2588                 return NULL;
2589
2590         page = vm_normal_page_pmd(vma, addr, pmd);
2591         if (!page)
2592                 return NULL;
2593
2594         if (PageReserved(page))
2595                 return NULL;
2596
2597         nid = page_to_nid(page);
2598         if (!node_isset(nid, node_states[N_MEMORY]))
2599                 return NULL;
2600
2601         return page;
2602 }
2603 #endif
2604
2605 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
2606                 unsigned long end, struct mm_walk *walk)
2607 {
2608         struct numa_maps *md = walk->private;
2609         struct vm_area_struct *vma = walk->vma;
2610         spinlock_t *ptl;
2611         pte_t *orig_pte;
2612         pte_t *pte;
2613
2614 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2615         ptl = pmd_trans_huge_lock(pmd, vma);
2616         if (ptl) {
2617                 struct page *page;
2618
2619                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
2620                 if (page)
2621                         gather_stats(page, md, pmd_dirty(*pmd),
2622                                      HPAGE_PMD_SIZE/PAGE_SIZE);
2623                 spin_unlock(ptl);
2624                 return 0;
2625         }
2626 #endif
2627         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
2628         if (!pte) {
2629                 walk->action = ACTION_AGAIN;
2630                 return 0;
2631         }
2632         do {
2633                 pte_t ptent = ptep_get(pte);
2634                 struct page *page = can_gather_numa_stats(ptent, vma, addr);
2635                 if (!page)
2636                         continue;
2637                 gather_stats(page, md, pte_dirty(ptent), 1);
2638
2639         } while (pte++, addr += PAGE_SIZE, addr != end);
2640         pte_unmap_unlock(orig_pte, ptl);
2641         cond_resched();
2642         return 0;
2643 }
2644 #ifdef CONFIG_HUGETLB_PAGE
2645 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
2646                 unsigned long addr, unsigned long end, struct mm_walk *walk)
2647 {
2648         pte_t huge_pte = huge_ptep_get(pte);
2649         struct numa_maps *md;
2650         struct page *page;
2651
2652         if (!pte_present(huge_pte))
2653                 return 0;
2654
2655         page = pte_page(huge_pte);
2656
2657         md = walk->private;
2658         gather_stats(page, md, pte_dirty(huge_pte), 1);
2659         return 0;
2660 }
2661
2662 #else
2663 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
2664                 unsigned long addr, unsigned long end, struct mm_walk *walk)
2665 {
2666         return 0;
2667 }
2668 #endif
2669
2670 static const struct mm_walk_ops show_numa_ops = {
2671         .hugetlb_entry = gather_hugetlb_stats,
2672         .pmd_entry = gather_pte_stats,
2673         .walk_lock = PGWALK_RDLOCK,
2674 };
2675
2676 /*
2677  * Display pages allocated per node and memory policy via /proc.
2678  */
2679 static int show_numa_map(struct seq_file *m, void *v)
2680 {
2681         struct numa_maps_private *numa_priv = m->private;
2682         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
2683         struct vm_area_struct *vma = v;
2684         struct numa_maps *md = &numa_priv->md;
2685         struct file *file = vma->vm_file;
2686         struct mm_struct *mm = vma->vm_mm;
2687         char buffer[64];
2688         struct mempolicy *pol;
2689         pgoff_t ilx;
2690         int nid;
2691
2692         if (!mm)
2693                 return 0;
2694
2695         /* Ensure we start with an empty set of numa_maps statistics. */
2696         memset(md, 0, sizeof(*md));
2697
2698         pol = __get_vma_policy(vma, vma->vm_start, &ilx);
2699         if (pol) {
2700                 mpol_to_str(buffer, sizeof(buffer), pol);
2701                 mpol_cond_put(pol);
2702         } else {
2703                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
2704         }
2705
2706         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
2707
2708         if (file) {
2709                 seq_puts(m, " file=");
2710                 seq_path(m, file_user_path(file), "\n\t= ");
2711         } else if (vma_is_initial_heap(vma)) {
2712                 seq_puts(m, " heap");
2713         } else if (vma_is_initial_stack(vma)) {
2714                 seq_puts(m, " stack");
2715         }
2716
2717         if (is_vm_hugetlb_page(vma))
2718                 seq_puts(m, " huge");
2719
2720         /* mmap_lock is held by m_start */
2721         walk_page_vma(vma, &show_numa_ops, md);
2722
2723         if (!md->pages)
2724                 goto out;
2725
2726         if (md->anon)
2727                 seq_printf(m, " anon=%lu", md->anon);
2728
2729         if (md->dirty)
2730                 seq_printf(m, " dirty=%lu", md->dirty);
2731
2732         if (md->pages != md->anon && md->pages != md->dirty)
2733                 seq_printf(m, " mapped=%lu", md->pages);
2734
2735         if (md->mapcount_max > 1)
2736                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
2737
2738         if (md->swapcache)
2739                 seq_printf(m, " swapcache=%lu", md->swapcache);
2740
2741         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2742                 seq_printf(m, " active=%lu", md->active);
2743
2744         if (md->writeback)
2745                 seq_printf(m, " writeback=%lu", md->writeback);
2746
2747         for_each_node_state(nid, N_MEMORY)
2748                 if (md->node[nid])
2749                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2750
2751         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2752 out:
2753         seq_putc(m, '\n');
2754         return 0;
2755 }
2756
2757 static const struct seq_operations proc_pid_numa_maps_op = {
2758         .start  = m_start,
2759         .next   = m_next,
2760         .stop   = m_stop,
2761         .show   = show_numa_map,
2762 };
2763
2764 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2765 {
2766         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2767                                 sizeof(struct numa_maps_private));
2768 }
2769
2770 const struct file_operations proc_pid_numa_maps_operations = {
2771         .open           = pid_numa_maps_open,
2772         .read           = seq_read,
2773         .llseek         = seq_lseek,
2774         .release        = proc_map_release,
2775 };
2776
2777 #endif /* CONFIG_NUMA */