fs/proc/task_mmu.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/pagewalk.h>
   3 #include <linux/mm_inline.h>
   4 #include <linux/hugetlb.h>
   5 #include <linux/huge_mm.h>
   6 #include <linux/mount.h>
   7 #include <linux/seq_file.h>
   8 #include <linux/highmem.h>
   9 #include <linux/ptrace.h>
  10 #include <linux/slab.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/mempolicy.h>
  13 #include <linux/rmap.h>
  14 #include <linux/swap.h>
  15 #include <linux/sched/mm.h>
  16 #include <linux/swapops.h>
  17 #include <linux/mmu_notifier.h>
  18 #include <linux/page_idle.h>
  19 #include <linux/shmem_fs.h>
  20 #include <linux/uaccess.h>
  21 #include <linux/pkeys.h>
  22
  23 #include <asm/elf.h>
  24 #include <asm/tlb.h>
  25 #include <asm/tlbflush.h>
  26 #include "internal.h"
  27
  28 #define SEQ_PUT_DEC(str, val) \
  29                 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
  30 void task_mem(struct seq_file *m, struct mm_struct *mm)
  31 {
  32         unsigned long text, lib, swap, anon, file, shmem;
  33         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  34
  35         anon = get_mm_counter(mm, MM_ANONPAGES);
  36         file = get_mm_counter(mm, MM_FILEPAGES);
  37         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  38
  39         /*
  40          * Note: to minimize their overhead, mm maintains hiwater_vm and
  41          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  42          * collector of these hiwater stats must therefore get total_vm
  43          * and rss too, which will usually be the higher.  Barriers? not
  44          * worth the effort, such snapshots can always be inconsistent.
  45          */
  46         hiwater_vm = total_vm = mm->total_vm;
  47         if (hiwater_vm < mm->hiwater_vm)
  48                 hiwater_vm = mm->hiwater_vm;
  49         hiwater_rss = total_rss = anon + file + shmem;
  50         if (hiwater_rss < mm->hiwater_rss)
  51                 hiwater_rss = mm->hiwater_rss;
  52
  53         /* split executable areas between text and lib */
  54         text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
  55         text = min(text, mm->exec_vm << PAGE_SHIFT);
  56         lib = (mm->exec_vm << PAGE_SHIFT) - text;
  57
  58         swap = get_mm_counter(mm, MM_SWAPENTS);
  59         SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
  60         SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
  61         SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
  62         SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
  63         SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
  64         SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
  65         SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
  66         SEQ_PUT_DEC(" kB\nRssFile:\t", file);
  67         SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
  68         SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
  69         SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
  70         seq_put_decimal_ull_width(m,
  71                     " kB\nVmExe:\t", text >> 10, 8);
  72         seq_put_decimal_ull_width(m,
  73                     " kB\nVmLib:\t", lib >> 10, 8);
  74         seq_put_decimal_ull_width(m,
  75                     " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
  76         SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
  77         seq_puts(m, " kB\n");
  78         hugetlb_report_usage(m, mm);
  79 }
  80 #undef SEQ_PUT_DEC
  81
  82 unsigned long task_vsize(struct mm_struct *mm)
  83 {
  84         return PAGE_SIZE * mm->total_vm;
  85 }
  86
  87 unsigned long task_statm(struct mm_struct *mm,
  88                          unsigned long *shared, unsigned long *text,
  89                          unsigned long *data, unsigned long *resident)
  90 {
  91         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  92                         get_mm_counter(mm, MM_SHMEMPAGES);
  93         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  94                                                                 >> PAGE_SHIFT;
  95         *data = mm->data_vm + mm->stack_vm;
  96         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
  97         return mm->total_vm;
  98 }
  99
 100 #ifdef CONFIG_NUMA
 101 /*
 102  * Save get_task_policy() for show_numa_map().
 103  */
 104 static void hold_task_mempolicy(struct proc_maps_private *priv)
 105 {
 106         struct task_struct *task = priv->task;
 107
 108         task_lock(task);
 109         priv->task_mempolicy = get_task_policy(task);
 110         mpol_get(priv->task_mempolicy);
 111         task_unlock(task);
 112 }
 113 static void release_task_mempolicy(struct proc_maps_private *priv)
 114 {
 115         mpol_put(priv->task_mempolicy);
 116 }
 117 #else
 118 static void hold_task_mempolicy(struct proc_maps_private *priv)
 119 {
 120 }
 121 static void release_task_mempolicy(struct proc_maps_private *priv)
 122 {
 123 }
 124 #endif
 125
 126 static struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv,
 127                                                 loff_t *ppos)
 128 {
 129         struct vm_area_struct *vma = vma_next(&priv->iter);
 130
 131         if (vma) {
 132                 *ppos = vma->vm_start;
 133         } else {
 134                 *ppos = -2UL;
 135                 vma = get_gate_vma(priv->mm);
 136         }
 137
 138         return vma;
 139 }
 140
 141 static void *m_start(struct seq_file *m, loff_t *ppos)
 142 {
 143         struct proc_maps_private *priv = m->private;
 144         unsigned long last_addr = *ppos;
 145         struct mm_struct *mm;
 146
 147         /* See m_next(). Zero at the start or after lseek. */
 148         if (last_addr == -1UL)
 149                 return NULL;
 150
 151         priv->task = get_proc_task(priv->inode);
 152         if (!priv->task)
 153                 return ERR_PTR(-ESRCH);
 154
 155         mm = priv->mm;
 156         if (!mm || !mmget_not_zero(mm)) {
 157                 put_task_struct(priv->task);
 158                 priv->task = NULL;
 159                 return NULL;
 160         }
 161
 162         if (mmap_read_lock_killable(mm)) {
 163                 mmput(mm);
 164                 put_task_struct(priv->task);
 165                 priv->task = NULL;
 166                 return ERR_PTR(-EINTR);
 167         }
 168
 169         vma_iter_init(&priv->iter, mm, last_addr);
 170         hold_task_mempolicy(priv);
 171         if (last_addr == -2UL)
 172                 return get_gate_vma(mm);
 173
 174         return proc_get_vma(priv, ppos);
 175 }
 176
 177 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
 178 {
 179         if (*ppos == -2UL) {
 180                 *ppos = -1UL;
 181                 return NULL;
 182         }
 183         return proc_get_vma(m->private, ppos);
 184 }
 185
 186 static void m_stop(struct seq_file *m, void *v)
 187 {
 188         struct proc_maps_private *priv = m->private;
 189         struct mm_struct *mm = priv->mm;
 190
 191         if (!priv->task)
 192                 return;
 193
 194         release_task_mempolicy(priv);
 195         mmap_read_unlock(mm);
 196         mmput(mm);
 197         put_task_struct(priv->task);
 198         priv->task = NULL;
 199 }
 200
 201 static int proc_maps_open(struct inode *inode, struct file *file,
 202                         const struct seq_operations *ops, int psize)
 203 {
 204         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 205
 206         if (!priv)
 207                 return -ENOMEM;
 208
 209         priv->inode = inode;
 210         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 211         if (IS_ERR(priv->mm)) {
 212                 int err = PTR_ERR(priv->mm);
 213
 214                 seq_release_private(inode, file);
 215                 return err;
 216         }
 217
 218         return 0;
 219 }
 220
 221 static int proc_map_release(struct inode *inode, struct file *file)
 222 {
 223         struct seq_file *seq = file->private_data;
 224         struct proc_maps_private *priv = seq->private;
 225
 226         if (priv->mm)
 227                 mmdrop(priv->mm);
 228
 229         return seq_release_private(inode, file);
 230 }
 231
 232 static int do_maps_open(struct inode *inode, struct file *file,
 233                         const struct seq_operations *ops)
 234 {
 235         return proc_maps_open(inode, file, ops,
 236                                 sizeof(struct proc_maps_private));
 237 }
 238
 239 /*
 240  * Indicate if the VMA is a stack for the given task; for
 241  * /proc/PID/maps that is the stack of the main task.
 242  */
 243 static int is_stack(struct vm_area_struct *vma)
 244 {
 245         /*
 246          * We make no effort to guess what a given thread considers to be
 247          * its "stack".  It's not even well-defined for programs written
 248          * languages like Go.
 249          */
 250         return vma->vm_start <= vma->vm_mm->start_stack &&
 251                 vma->vm_end >= vma->vm_mm->start_stack;
 252 }
 253
 254 static void show_vma_header_prefix(struct seq_file *m,
 255                                    unsigned long start, unsigned long end,
 256                                    vm_flags_t flags, unsigned long long pgoff,
 257                                    dev_t dev, unsigned long ino)
 258 {
 259         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 260         seq_put_hex_ll(m, NULL, start, 8);
 261         seq_put_hex_ll(m, "-", end, 8);
 262         seq_putc(m, ' ');
 263         seq_putc(m, flags & VM_READ ? 'r' : '-');
 264         seq_putc(m, flags & VM_WRITE ? 'w' : '-');
 265         seq_putc(m, flags & VM_EXEC ? 'x' : '-');
 266         seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
 267         seq_put_hex_ll(m, " ", pgoff, 8);
 268         seq_put_hex_ll(m, " ", MAJOR(dev), 2);
 269         seq_put_hex_ll(m, ":", MINOR(dev), 2);
 270         seq_put_decimal_ull(m, " ", ino);
 271         seq_putc(m, ' ');
 272 }
 273
 274 static void
 275 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 276 {
 277         struct mm_struct *mm = vma->vm_mm;
 278         struct file *file = vma->vm_file;
 279         vm_flags_t flags = vma->vm_flags;
 280         unsigned long ino = 0;
 281         unsigned long long pgoff = 0;
 282         unsigned long start, end;
 283         dev_t dev = 0;
 284         const char *name = NULL;
 285
 286         if (file) {
 287                 struct inode *inode = file_inode(vma->vm_file);
 288                 dev = inode->i_sb->s_dev;
 289                 ino = inode->i_ino;
 290                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 291         }
 292
 293         start = vma->vm_start;
 294         end = vma->vm_end;
 295         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 296
 297         /*
 298          * Print the dentry name for named mappings, and a
 299          * special [heap] marker for the heap:
 300          */
 301         if (file) {
 302                 seq_pad(m, ' ');
 303                 seq_file_path(m, file, "\n");
 304                 goto done;
 305         }
 306
 307         if (vma->vm_ops && vma->vm_ops->name) {
 308                 name = vma->vm_ops->name(vma);
 309                 if (name)
 310                         goto done;
 311         }
 312
 313         name = arch_vma_name(vma);
 314         if (!name) {
 315                 struct anon_vma_name *anon_name;
 316
 317                 if (!mm) {
 318                         name = "[vdso]";
 319                         goto done;
 320                 }
 321
 322                 if (vma->vm_start <= mm->brk &&
 323                     vma->vm_end >= mm->start_brk) {
 324                         name = "[heap]";
 325                         goto done;
 326                 }
 327
 328                 if (is_stack(vma)) {
 329                         name = "[stack]";
 330                         goto done;
 331                 }
 332
 333                 anon_name = anon_vma_name(vma);
 334                 if (anon_name) {
 335                         seq_pad(m, ' ');
 336                         seq_printf(m, "[anon:%s]", anon_name->name);
 337                 }
 338         }
 339
 340 done:
 341         if (name) {
 342                 seq_pad(m, ' ');
 343                 seq_puts(m, name);
 344         }
 345         seq_putc(m, '\n');
 346 }
 347
 348 static int show_map(struct seq_file *m, void *v)
 349 {
 350         show_map_vma(m, v);
 351         return 0;
 352 }
 353
 354 static const struct seq_operations proc_pid_maps_op = {
 355         .start  = m_start,
 356         .next   = m_next,
 357         .stop   = m_stop,
 358         .show   = show_map
 359 };
 360
 361 static int pid_maps_open(struct inode *inode, struct file *file)
 362 {
 363         return do_maps_open(inode, file, &proc_pid_maps_op);
 364 }
 365
 366 const struct file_operations proc_pid_maps_operations = {
 367         .open           = pid_maps_open,
 368         .read           = seq_read,
 369         .llseek         = seq_lseek,
 370         .release        = proc_map_release,
 371 };
 372
 373 /*
 374  * Proportional Set Size(PSS): my share of RSS.
 375  *
 376  * PSS of a process is the count of pages it has in memory, where each
 377  * page is divided by the number of processes sharing it.  So if a
 378  * process has 1000 pages all to itself, and 1000 shared with one other
 379  * process, its PSS will be 1500.
 380  *
 381  * To keep (accumulated) division errors low, we adopt a 64bit
 382  * fixed-point pss counter to minimize division errors. So (pss >>
 383  * PSS_SHIFT) would be the real byte count.
 384  *
 385  * A shift of 12 before division means (assuming 4K page size):
 386  *      - 1M 3-user-pages add up to 8KB errors;
 387  *      - supports mapcount up to 2^24, or 16M;
 388  *      - supports PSS up to 2^52 bytes, or 4PB.
 389  */
 390 #define PSS_SHIFT 12
 391
 392 #ifdef CONFIG_PROC_PAGE_MONITOR
 393 struct mem_size_stats {
 394         unsigned long resident;
 395         unsigned long shared_clean;
 396         unsigned long shared_dirty;
 397         unsigned long private_clean;
 398         unsigned long private_dirty;
 399         unsigned long referenced;
 400         unsigned long anonymous;
 401         unsigned long lazyfree;
 402         unsigned long anonymous_thp;
 403         unsigned long shmem_thp;
 404         unsigned long file_thp;
 405         unsigned long swap;
 406         unsigned long shared_hugetlb;
 407         unsigned long private_hugetlb;
 408         u64 pss;
 409         u64 pss_anon;
 410         u64 pss_file;
 411         u64 pss_shmem;
 412         u64 pss_dirty;
 413         u64 pss_locked;
 414         u64 swap_pss;
 415 };
 416
 417 static void smaps_page_accumulate(struct mem_size_stats *mss,
 418                 struct page *page, unsigned long size, unsigned long pss,
 419                 bool dirty, bool locked, bool private)
 420 {
 421         mss->pss += pss;
 422
 423         if (PageAnon(page))
 424                 mss->pss_anon += pss;
 425         else if (PageSwapBacked(page))
 426                 mss->pss_shmem += pss;
 427         else
 428                 mss->pss_file += pss;
 429
 430         if (locked)
 431                 mss->pss_locked += pss;
 432
 433         if (dirty || PageDirty(page)) {
 434                 mss->pss_dirty += pss;
 435                 if (private)
 436                         mss->private_dirty += size;
 437                 else
 438                         mss->shared_dirty += size;
 439         } else {
 440                 if (private)
 441                         mss->private_clean += size;
 442                 else
 443                         mss->shared_clean += size;
 444         }
 445 }
 446
 447 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 448                 bool compound, bool young, bool dirty, bool locked,
 449                 bool migration)
 450 {
 451         int i, nr = compound ? compound_nr(page) : 1;
 452         unsigned long size = nr * PAGE_SIZE;
 453
 454         /*
 455          * First accumulate quantities that depend only on |size| and the type
 456          * of the compound page.
 457          */
 458         if (PageAnon(page)) {
 459                 mss->anonymous += size;
 460                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 461                         mss->lazyfree += size;
 462         }
 463
 464         mss->resident += size;
 465         /* Accumulate the size in pages that have been accessed. */
 466         if (young || page_is_young(page) || PageReferenced(page))
 467                 mss->referenced += size;
 468
 469         /*
 470          * Then accumulate quantities that may depend on sharing, or that may
 471          * differ page-by-page.
 472          *
 473          * page_count(page) == 1 guarantees the page is mapped exactly once.
 474          * If any subpage of the compound page mapped with PTE it would elevate
 475          * page_count().
 476          *
 477          * The page_mapcount() is called to get a snapshot of the mapcount.
 478          * Without holding the page lock this snapshot can be slightly wrong as
 479          * we cannot always read the mapcount atomically.  It is not safe to
 480          * call page_mapcount() even with PTL held if the page is not mapped,
 481          * especially for migration entries.  Treat regular migration entries
 482          * as mapcount == 1.
 483          */
 484         if ((page_count(page) == 1) || migration) {
 485                 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
 486                         locked, true);
 487                 return;
 488         }
 489         for (i = 0; i < nr; i++, page++) {
 490                 int mapcount = page_mapcount(page);
 491                 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
 492                 if (mapcount >= 2)
 493                         pss /= mapcount;
 494                 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
 495                                       mapcount < 2);
 496         }
 497 }
 498
 499 #ifdef CONFIG_SHMEM
 500 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 501                           __always_unused int depth, struct mm_walk *walk)
 502 {
 503         struct mem_size_stats *mss = walk->private;
 504         struct vm_area_struct *vma = walk->vma;
 505
 506         mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping,
 507                                               linear_page_index(vma, addr),
 508                                               linear_page_index(vma, end));
 509
 510         return 0;
 511 }
 512 #else
 513 #define smaps_pte_hole          NULL
 514 #endif /* CONFIG_SHMEM */
 515
 516 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
 517 {
 518 #ifdef CONFIG_SHMEM
 519         if (walk->ops->pte_hole) {
 520                 /* depth is not used */
 521                 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
 522         }
 523 #endif
 524 }
 525
 526 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 527                 struct mm_walk *walk)
 528 {
 529         struct mem_size_stats *mss = walk->private;
 530         struct vm_area_struct *vma = walk->vma;
 531         bool locked = !!(vma->vm_flags & VM_LOCKED);
 532         struct page *page = NULL;
 533         bool migration = false, young = false, dirty = false;
 534
 535         if (pte_present(*pte)) {
 536                 page = vm_normal_page(vma, addr, *pte);
 537                 young = pte_young(*pte);
 538                 dirty = pte_dirty(*pte);
 539         } else if (is_swap_pte(*pte)) {
 540                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 541
 542                 if (!non_swap_entry(swpent)) {
 543                         int mapcount;
 544
 545                         mss->swap += PAGE_SIZE;
 546                         mapcount = swp_swapcount(swpent);
 547                         if (mapcount >= 2) {
 548                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 549
 550                                 do_div(pss_delta, mapcount);
 551                                 mss->swap_pss += pss_delta;
 552                         } else {
 553                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 554                         }
 555                 } else if (is_pfn_swap_entry(swpent)) {
 556                         if (is_migration_entry(swpent))
 557                                 migration = true;
 558                         page = pfn_swap_entry_to_page(swpent);
 559                 }
 560         } else {
 561                 smaps_pte_hole_lookup(addr, walk);
 562                 return;
 563         }
 564
 565         if (!page)
 566                 return;
 567
 568         smaps_account(mss, page, false, young, dirty, locked, migration);
 569 }
 570
 571 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 572 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 573                 struct mm_walk *walk)
 574 {
 575         struct mem_size_stats *mss = walk->private;
 576         struct vm_area_struct *vma = walk->vma;
 577         bool locked = !!(vma->vm_flags & VM_LOCKED);
 578         struct page *page = NULL;
 579         bool migration = false;
 580
 581         if (pmd_present(*pmd)) {
 582                 /* FOLL_DUMP will return -EFAULT on huge zero page */
 583                 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 584         } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
 585                 swp_entry_t entry = pmd_to_swp_entry(*pmd);
 586
 587                 if (is_migration_entry(entry)) {
 588                         migration = true;
 589                         page = pfn_swap_entry_to_page(entry);
 590                 }
 591         }
 592         if (IS_ERR_OR_NULL(page))
 593                 return;
 594         if (PageAnon(page))
 595                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 596         else if (PageSwapBacked(page))
 597                 mss->shmem_thp += HPAGE_PMD_SIZE;
 598         else if (is_zone_device_page(page))
 599                 /* pass */;
 600         else
 601                 mss->file_thp += HPAGE_PMD_SIZE;
 602
 603         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
 604                       locked, migration);
 605 }
 606 #else
 607 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 608                 struct mm_walk *walk)
 609 {
 610 }
 611 #endif
 612
 613 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 614                            struct mm_walk *walk)
 615 {
 616         struct vm_area_struct *vma = walk->vma;
 617         pte_t *pte;
 618         spinlock_t *ptl;
 619
 620         ptl = pmd_trans_huge_lock(pmd, vma);
 621         if (ptl) {
 622                 smaps_pmd_entry(pmd, addr, walk);
 623                 spin_unlock(ptl);
 624                 goto out;
 625         }
 626
 627         if (pmd_trans_unstable(pmd))
 628                 goto out;
 629         /*
 630          * The mmap_lock held all the way back in m_start() is what
 631          * keeps khugepaged out of here and from collapsing things
 632          * in here.
 633          */
 634         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 635         for (; addr != end; pte++, addr += PAGE_SIZE)
 636                 smaps_pte_entry(pte, addr, walk);
 637         pte_unmap_unlock(pte - 1, ptl);
 638 out:
 639         cond_resched();
 640         return 0;
 641 }
 642
 643 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 644 {
 645         /*
 646          * Don't forget to update Documentation/ on changes.
 647          */
 648         static const char mnemonics[BITS_PER_LONG][2] = {
 649                 /*
 650                  * In case if we meet a flag we don't know about.
 651                  */
 652                 [0 ... (BITS_PER_LONG-1)] = "??",
 653
 654                 [ilog2(VM_READ)]        = "rd",
 655                 [ilog2(VM_WRITE)]       = "wr",
 656                 [ilog2(VM_EXEC)]        = "ex",
 657                 [ilog2(VM_SHARED)]      = "sh",
 658                 [ilog2(VM_MAYREAD)]     = "mr",
 659                 [ilog2(VM_MAYWRITE)]    = "mw",
 660                 [ilog2(VM_MAYEXEC)]     = "me",
 661                 [ilog2(VM_MAYSHARE)]    = "ms",
 662                 [ilog2(VM_GROWSDOWN)]   = "gd",
 663                 [ilog2(VM_PFNMAP)]      = "pf",
 664                 [ilog2(VM_LOCKED)]      = "lo",
 665                 [ilog2(VM_IO)]          = "io",
 666                 [ilog2(VM_SEQ_READ)]    = "sr",
 667                 [ilog2(VM_RAND_READ)]   = "rr",
 668                 [ilog2(VM_DONTCOPY)]    = "dc",
 669                 [ilog2(VM_DONTEXPAND)]  = "de",
 670                 [ilog2(VM_ACCOUNT)]     = "ac",
 671                 [ilog2(VM_NORESERVE)]   = "nr",
 672                 [ilog2(VM_HUGETLB)]     = "ht",
 673                 [ilog2(VM_SYNC)]        = "sf",
 674                 [ilog2(VM_ARCH_1)]      = "ar",
 675                 [ilog2(VM_WIPEONFORK)]  = "wf",
 676                 [ilog2(VM_DONTDUMP)]    = "dd",
 677 #ifdef CONFIG_ARM64_BTI
 678                 [ilog2(VM_ARM64_BTI)]   = "bt",
 679 #endif
 680 #ifdef CONFIG_MEM_SOFT_DIRTY
 681                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 682 #endif
 683                 [ilog2(VM_MIXEDMAP)]    = "mm",
 684                 [ilog2(VM_HUGEPAGE)]    = "hg",
 685                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 686                 [ilog2(VM_MERGEABLE)]   = "mg",
 687                 [ilog2(VM_UFFD_MISSING)]= "um",
 688                 [ilog2(VM_UFFD_WP)]     = "uw",
 689 #ifdef CONFIG_ARM64_MTE
 690                 [ilog2(VM_MTE)]         = "mt",
 691                 [ilog2(VM_MTE_ALLOWED)] = "",
 692 #endif
 693 #ifdef CONFIG_ARCH_HAS_PKEYS
 694                 /* These come out via ProtectionKey: */
 695                 [ilog2(VM_PKEY_BIT0)]   = "",
 696                 [ilog2(VM_PKEY_BIT1)]   = "",
 697                 [ilog2(VM_PKEY_BIT2)]   = "",
 698                 [ilog2(VM_PKEY_BIT3)]   = "",
 699 #if VM_PKEY_BIT4
 700                 [ilog2(VM_PKEY_BIT4)]   = "",
 701 #endif
 702 #endif /* CONFIG_ARCH_HAS_PKEYS */
 703 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
 704                 [ilog2(VM_UFFD_MINOR)]  = "ui",
 705 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
 706         };
 707         size_t i;
 708
 709         seq_puts(m, "VmFlags: ");
 710         for (i = 0; i < BITS_PER_LONG; i++) {
 711                 if (!mnemonics[i][0])
 712                         continue;
 713                 if (vma->vm_flags & (1UL << i)) {
 714                         seq_putc(m, mnemonics[i][0]);
 715                         seq_putc(m, mnemonics[i][1]);
 716                         seq_putc(m, ' ');
 717                 }
 718         }
 719         seq_putc(m, '\n');
 720 }
 721
 722 #ifdef CONFIG_HUGETLB_PAGE
 723 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 724                                  unsigned long addr, unsigned long end,
 725                                  struct mm_walk *walk)
 726 {
 727         struct mem_size_stats *mss = walk->private;
 728         struct vm_area_struct *vma = walk->vma;
 729         struct page *page = NULL;
 730
 731         if (pte_present(*pte)) {
 732                 page = vm_normal_page(vma, addr, *pte);
 733         } else if (is_swap_pte(*pte)) {
 734                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 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 };
 755
 756 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 757         .pmd_entry              = smaps_pte_range,
 758         .hugetlb_entry          = smaps_hugetlb_range,
 759         .pte_hole               = smaps_pte_hole,
 760 };
 761
 762 /*
 763  * Gather mem stats from @vma with the indicated beginning
 764  * address @start, and keep them in @mss.
 765  *
 766  * Use vm_start of @vma as the beginning address if @start is 0.
 767  */
 768 static void smap_gather_stats(struct vm_area_struct *vma,
 769                 struct mem_size_stats *mss, unsigned long start)
 770 {
 771         const struct mm_walk_ops *ops = &smaps_walk_ops;
 772
 773         /* Invalid start */
 774         if (start >= vma->vm_end)
 775                 return;
 776
 777 #ifdef CONFIG_SHMEM
 778         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 779                 /*
 780                  * For shared or readonly shmem mappings we know that all
 781                  * swapped out pages belong to the shmem object, and we can
 782                  * obtain the swap value much more efficiently. For private
 783                  * writable mappings, we might have COW pages that are
 784                  * not affected by the parent swapped out pages of the shmem
 785                  * object, so we have to distinguish them during the page walk.
 786                  * Unless we know that the shmem object (or the part mapped by
 787                  * our VMA) has no swapped out pages at all.
 788                  */
 789                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 790
 791                 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 792                                         !(vma->vm_flags & VM_WRITE))) {
 793                         mss->swap += shmem_swapped;
 794                 } else {
 795                         ops = &smaps_shmem_walk_ops;
 796                 }
 797         }
 798 #endif
 799         /* mmap_lock is held in m_start */
 800         if (!start)
 801                 walk_page_vma(vma, ops, mss);
 802         else
 803                 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
 804 }
 805
 806 #define SEQ_PUT_DEC(str, val) \
 807                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 808
 809 /* Show the contents common for smaps and smaps_rollup */
 810 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 811         bool rollup_mode)
 812 {
 813         SEQ_PUT_DEC("Rss:            ", mss->resident);
 814         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 815         SEQ_PUT_DEC(" kB\nPss_Dirty:      ", mss->pss_dirty >> PSS_SHIFT);
 816         if (rollup_mode) {
 817                 /*
 818                  * These are meaningful only for smaps_rollup, otherwise two of
 819                  * them are zero, and the other one is the same as Pss.
 820                  */
 821                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 822                         mss->pss_anon >> PSS_SHIFT);
 823                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 824                         mss->pss_file >> PSS_SHIFT);
 825                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 826                         mss->pss_shmem >> PSS_SHIFT);
 827         }
 828         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 829         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 830         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 831         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 832         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 833         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 834         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 835         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 836         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 837         SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
 838         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 839         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 840                                   mss->private_hugetlb >> 10, 7);
 841         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 842         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 843                                         mss->swap_pss >> PSS_SHIFT);
 844         SEQ_PUT_DEC(" kB\nLocked:         ",
 845                                         mss->pss_locked >> PSS_SHIFT);
 846         seq_puts(m, " kB\n");
 847 }
 848
 849 static int show_smap(struct seq_file *m, void *v)
 850 {
 851         struct vm_area_struct *vma = v;
 852         struct mem_size_stats mss;
 853
 854         memset(&mss, 0, sizeof(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:    %d\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         MA_STATE(mas, &mm->mm_mt, 0, 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         memset(&mss, 0, sizeof(mss));
 897
 898         ret = mmap_read_lock_killable(mm);
 899         if (ret)
 900                 goto out_put_mm;
 901
 902         hold_task_mempolicy(priv);
 903         vma = mas_find(&mas, ULONG_MAX);
 904
 905         if (unlikely(!vma))
 906                 goto empty_set;
 907
 908         vma_start = vma->vm_start;
 909         do {
 910                 smap_gather_stats(vma, &mss, 0);
 911                 last_vma_end = vma->vm_end;
 912
 913                 /*
 914                  * Release mmap_lock temporarily if someone wants to
 915                  * access it for write request.
 916                  */
 917                 if (mmap_lock_is_contended(mm)) {
 918                         mas_pause(&mas);
 919                         mmap_read_unlock(mm);
 920                         ret = mmap_read_lock_killable(mm);
 921                         if (ret) {
 922                                 release_task_mempolicy(priv);
 923                                 goto out_put_mm;
 924                         }
 925
 926                         /*
 927                          * After dropping the lock, there are four cases to
 928                          * consider. See the following example for explanation.
 929                          *
 930                          *   +------+------+-----------+
 931                          *   | VMA1 | VMA2 | VMA3      |
 932                          *   +------+------+-----------+
 933                          *   |      |      |           |
 934                          *  4k     8k     16k         400k
 935                          *
 936                          * Suppose we drop the lock after reading VMA2 due to
 937                          * contention, then we get:
 938                          *
 939                          *      last_vma_end = 16k
 940                          *
 941                          * 1) VMA2 is freed, but VMA3 exists:
 942                          *
 943                          *    find_vma(mm, 16k - 1) will return VMA3.
 944                          *    In this case, just continue from VMA3.
 945                          *
 946                          * 2) VMA2 still exists:
 947                          *
 948                          *    find_vma(mm, 16k - 1) will return VMA2.
 949                          *    Iterate the loop like the original one.
 950                          *
 951                          * 3) No more VMAs can be found:
 952                          *
 953                          *    find_vma(mm, 16k - 1) will return NULL.
 954                          *    No more things to do, just break.
 955                          *
 956                          * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
 957                          *
 958                          *    find_vma(mm, 16k - 1) will return VMA' whose range
 959                          *    contains last_vma_end.
 960                          *    Iterate VMA' from last_vma_end.
 961                          */
 962                         vma = mas_find(&mas, ULONG_MAX);
 963                         /* Case 3 above */
 964                         if (!vma)
 965                                 break;
 966
 967                         /* Case 1 above */
 968                         if (vma->vm_start >= last_vma_end)
 969                                 continue;
 970
 971                         /* Case 4 above */
 972                         if (vma->vm_end > last_vma_end)
 973                                 smap_gather_stats(vma, &mss, last_vma_end);
 974                 }
 975                 /* Case 2 above */
 976         } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL);
 977
 978 empty_set:
 979         show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
 980         seq_pad(m, ' ');
 981         seq_puts(m, "[rollup]\n");
 982
 983         __show_smap(m, &mss, true);
 984
 985         release_task_mempolicy(priv);
 986         mmap_read_unlock(mm);
 987
 988 out_put_mm:
 989         mmput(mm);
 990 out_put_task:
 991         put_task_struct(priv->task);
 992         priv->task = NULL;
 993
 994         return ret;
 995 }
 996 #undef SEQ_PUT_DEC
 997
 998 static const struct seq_operations proc_pid_smaps_op = {
 999         .start  = m_start,
1000         .next   = m_next,
1001         .stop   = m_stop,
1002         .show   = show_smap
1003 };
1004
1005 static int pid_smaps_open(struct inode *inode, struct file *file)
1006 {
1007         return do_maps_open(inode, file, &proc_pid_smaps_op);
1008 }
1009
1010 static int smaps_rollup_open(struct inode *inode, struct file *file)
1011 {
1012         int ret;
1013         struct proc_maps_private *priv;
1014
1015         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1016         if (!priv)
1017                 return -ENOMEM;
1018
1019         ret = single_open(file, show_smaps_rollup, priv);
1020         if (ret)
1021                 goto out_free;
1022
1023         priv->inode = inode;
1024         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1025         if (IS_ERR(priv->mm)) {
1026                 ret = PTR_ERR(priv->mm);
1027
1028                 single_release(inode, file);
1029                 goto out_free;
1030         }
1031
1032         return 0;
1033
1034 out_free:
1035         kfree(priv);
1036         return ret;
1037 }
1038
1039 static int smaps_rollup_release(struct inode *inode, struct file *file)
1040 {
1041         struct seq_file *seq = file->private_data;
1042         struct proc_maps_private *priv = seq->private;
1043
1044         if (priv->mm)
1045                 mmdrop(priv->mm);
1046
1047         kfree(priv);
1048         return single_release(inode, file);
1049 }
1050
1051 const struct file_operations proc_pid_smaps_operations = {
1052         .open           = pid_smaps_open,
1053         .read           = seq_read,
1054         .llseek         = seq_lseek,
1055         .release        = proc_map_release,
1056 };
1057
1058 const struct file_operations proc_pid_smaps_rollup_operations = {
1059         .open           = smaps_rollup_open,
1060         .read           = seq_read,
1061         .llseek         = seq_lseek,
1062         .release        = smaps_rollup_release,
1063 };
1064
1065 enum clear_refs_types {
1066         CLEAR_REFS_ALL = 1,
1067         CLEAR_REFS_ANON,
1068         CLEAR_REFS_MAPPED,
1069         CLEAR_REFS_SOFT_DIRTY,
1070         CLEAR_REFS_MM_HIWATER_RSS,
1071         CLEAR_REFS_LAST,
1072 };
1073
1074 struct clear_refs_private {
1075         enum clear_refs_types type;
1076 };
1077
1078 #ifdef CONFIG_MEM_SOFT_DIRTY
1079
1080 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1081 {
1082         struct page *page;
1083
1084         if (!pte_write(pte))
1085                 return false;
1086         if (!is_cow_mapping(vma->vm_flags))
1087                 return false;
1088         if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1089                 return false;
1090         page = vm_normal_page(vma, addr, pte);
1091         if (!page)
1092                 return false;
1093         return page_maybe_dma_pinned(page);
1094 }
1095
1096 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1097                 unsigned long addr, pte_t *pte)
1098 {
1099         /*
1100          * The soft-dirty tracker uses #PF-s to catch writes
1101          * to pages, so write-protect the pte as well. See the
1102          * Documentation/admin-guide/mm/soft-dirty.rst for full description
1103          * of how soft-dirty works.
1104          */
1105         pte_t ptent = *pte;
1106
1107         if (pte_present(ptent)) {
1108                 pte_t old_pte;
1109
1110                 if (pte_is_pinned(vma, addr, ptent))
1111                         return;
1112                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1113                 ptent = pte_wrprotect(old_pte);
1114                 ptent = pte_clear_soft_dirty(ptent);
1115                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1116         } else if (is_swap_pte(ptent)) {
1117                 ptent = pte_swp_clear_soft_dirty(ptent);
1118                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1119         }
1120 }
1121 #else
1122 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1123                 unsigned long addr, pte_t *pte)
1124 {
1125 }
1126 #endif
1127
1128 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1129 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1130                 unsigned long addr, pmd_t *pmdp)
1131 {
1132         pmd_t old, pmd = *pmdp;
1133
1134         if (pmd_present(pmd)) {
1135                 /* See comment in change_huge_pmd() */
1136                 old = pmdp_invalidate(vma, addr, pmdp);
1137                 if (pmd_dirty(old))
1138                         pmd = pmd_mkdirty(pmd);
1139                 if (pmd_young(old))
1140                         pmd = pmd_mkyoung(pmd);
1141
1142                 pmd = pmd_wrprotect(pmd);
1143                 pmd = pmd_clear_soft_dirty(pmd);
1144
1145                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1146         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1147                 pmd = pmd_swp_clear_soft_dirty(pmd);
1148                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1149         }
1150 }
1151 #else
1152 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1153                 unsigned long addr, pmd_t *pmdp)
1154 {
1155 }
1156 #endif
1157
1158 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1159                                 unsigned long end, struct mm_walk *walk)
1160 {
1161         struct clear_refs_private *cp = walk->private;
1162         struct vm_area_struct *vma = walk->vma;
1163         pte_t *pte, ptent;
1164         spinlock_t *ptl;
1165         struct page *page;
1166
1167         ptl = pmd_trans_huge_lock(pmd, vma);
1168         if (ptl) {
1169                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1170                         clear_soft_dirty_pmd(vma, addr, pmd);
1171                         goto out;
1172                 }
1173
1174                 if (!pmd_present(*pmd))
1175                         goto out;
1176
1177                 page = pmd_page(*pmd);
1178
1179                 /* Clear accessed and referenced bits. */
1180                 pmdp_test_and_clear_young(vma, addr, pmd);
1181                 test_and_clear_page_young(page);
1182                 ClearPageReferenced(page);
1183 out:
1184                 spin_unlock(ptl);
1185                 return 0;
1186         }
1187
1188         if (pmd_trans_unstable(pmd))
1189                 return 0;
1190
1191         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1192         for (; addr != end; pte++, addr += PAGE_SIZE) {
1193                 ptent = *pte;
1194
1195                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1196                         clear_soft_dirty(vma, addr, pte);
1197                         continue;
1198                 }
1199
1200                 if (!pte_present(ptent))
1201                         continue;
1202
1203                 page = vm_normal_page(vma, addr, ptent);
1204                 if (!page)
1205                         continue;
1206
1207                 /* Clear accessed and referenced bits. */
1208                 ptep_test_and_clear_young(vma, addr, pte);
1209                 test_and_clear_page_young(page);
1210                 ClearPageReferenced(page);
1211         }
1212         pte_unmap_unlock(pte - 1, ptl);
1213         cond_resched();
1214         return 0;
1215 }
1216
1217 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1218                                 struct mm_walk *walk)
1219 {
1220         struct clear_refs_private *cp = walk->private;
1221         struct vm_area_struct *vma = walk->vma;
1222
1223         if (vma->vm_flags & VM_PFNMAP)
1224                 return 1;
1225
1226         /*
1227          * Writing 1 to /proc/pid/clear_refs affects all pages.
1228          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1229          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1230          * Writing 4 to /proc/pid/clear_refs affects all pages.
1231          */
1232         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1233                 return 1;
1234         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1235                 return 1;
1236         return 0;
1237 }
1238
1239 static const struct mm_walk_ops clear_refs_walk_ops = {
1240         .pmd_entry              = clear_refs_pte_range,
1241         .test_walk              = clear_refs_test_walk,
1242 };
1243
1244 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1245                                 size_t count, loff_t *ppos)
1246 {
1247         struct task_struct *task;
1248         char buffer[PROC_NUMBUF];
1249         struct mm_struct *mm;
1250         struct vm_area_struct *vma;
1251         enum clear_refs_types type;
1252         int itype;
1253         int rv;
1254
1255         memset(buffer, 0, sizeof(buffer));
1256         if (count > sizeof(buffer) - 1)
1257                 count = sizeof(buffer) - 1;
1258         if (copy_from_user(buffer, buf, count))
1259                 return -EFAULT;
1260         rv = kstrtoint(strstrip(buffer), 10, &itype);
1261         if (rv < 0)
1262                 return rv;
1263         type = (enum clear_refs_types)itype;
1264         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1265                 return -EINVAL;
1266
1267         task = get_proc_task(file_inode(file));
1268         if (!task)
1269                 return -ESRCH;
1270         mm = get_task_mm(task);
1271         if (mm) {
1272                 MA_STATE(mas, &mm->mm_mt, 0, 0);
1273                 struct mmu_notifier_range range;
1274                 struct clear_refs_private cp = {
1275                         .type = type,
1276                 };
1277
1278                 if (mmap_write_lock_killable(mm)) {
1279                         count = -EINTR;
1280                         goto out_mm;
1281                 }
1282                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1283                         /*
1284                          * Writing 5 to /proc/pid/clear_refs resets the peak
1285                          * resident set size to this mm's current rss value.
1286                          */
1287                         reset_mm_hiwater_rss(mm);
1288                         goto out_unlock;
1289                 }
1290
1291                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1292                         mas_for_each(&mas, vma, ULONG_MAX) {
1293                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1294                                         continue;
1295                                 vma->vm_flags &= ~VM_SOFTDIRTY;
1296                                 vma_set_page_prot(vma);
1297                         }
1298
1299                         inc_tlb_flush_pending(mm);
1300                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1301                                                 0, NULL, mm, 0, -1UL);
1302                         mmu_notifier_invalidate_range_start(&range);
1303                 }
1304                 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1305                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1306                         mmu_notifier_invalidate_range_end(&range);
1307                         flush_tlb_mm(mm);
1308                         dec_tlb_flush_pending(mm);
1309                 }
1310 out_unlock:
1311                 mmap_write_unlock(mm);
1312 out_mm:
1313                 mmput(mm);
1314         }
1315         put_task_struct(task);
1316
1317         return count;
1318 }
1319
1320 const struct file_operations proc_clear_refs_operations = {
1321         .write          = clear_refs_write,
1322         .llseek         = noop_llseek,
1323 };
1324
1325 typedef struct {
1326         u64 pme;
1327 } pagemap_entry_t;
1328
1329 struct pagemapread {
1330         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1331         pagemap_entry_t *buffer;
1332         bool show_pfn;
1333 };
1334
1335 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1336 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1337
1338 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1339 #define PM_PFRAME_BITS          55
1340 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1341 #define PM_SOFT_DIRTY           BIT_ULL(55)
1342 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1343 #define PM_UFFD_WP              BIT_ULL(57)
1344 #define PM_FILE                 BIT_ULL(61)
1345 #define PM_SWAP                 BIT_ULL(62)
1346 #define PM_PRESENT              BIT_ULL(63)
1347
1348 #define PM_END_OF_BUFFER    1
1349
1350 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1351 {
1352         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1353 }
1354
1355 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1356                           struct pagemapread *pm)
1357 {
1358         pm->buffer[pm->pos++] = *pme;
1359         if (pm->pos >= pm->len)
1360                 return PM_END_OF_BUFFER;
1361         return 0;
1362 }
1363
1364 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1365                             __always_unused int depth, struct mm_walk *walk)
1366 {
1367         struct pagemapread *pm = walk->private;
1368         unsigned long addr = start;
1369         int err = 0;
1370
1371         while (addr < end) {
1372                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1373                 pagemap_entry_t pme = make_pme(0, 0);
1374                 /* End of address space hole, which we mark as non-present. */
1375                 unsigned long hole_end;
1376
1377                 if (vma)
1378                         hole_end = min(end, vma->vm_start);
1379                 else
1380                         hole_end = end;
1381
1382                 for (; addr < hole_end; addr += PAGE_SIZE) {
1383                         err = add_to_pagemap(addr, &pme, pm);
1384                         if (err)
1385                                 goto out;
1386                 }
1387
1388                 if (!vma)
1389                         break;
1390
1391                 /* Addresses in the VMA. */
1392                 if (vma->vm_flags & VM_SOFTDIRTY)
1393                         pme = make_pme(0, PM_SOFT_DIRTY);
1394                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1395                         err = add_to_pagemap(addr, &pme, pm);
1396                         if (err)
1397                                 goto out;
1398                 }
1399         }
1400 out:
1401         return err;
1402 }
1403
1404 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1405                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1406 {
1407         u64 frame = 0, flags = 0;
1408         struct page *page = NULL;
1409         bool migration = false;
1410
1411         if (pte_present(pte)) {
1412                 if (pm->show_pfn)
1413                         frame = pte_pfn(pte);
1414                 flags |= PM_PRESENT;
1415                 page = vm_normal_page(vma, addr, pte);
1416                 if (pte_soft_dirty(pte))
1417                         flags |= PM_SOFT_DIRTY;
1418                 if (pte_uffd_wp(pte))
1419                         flags |= PM_UFFD_WP;
1420         } else if (is_swap_pte(pte)) {
1421                 swp_entry_t entry;
1422                 if (pte_swp_soft_dirty(pte))
1423                         flags |= PM_SOFT_DIRTY;
1424                 if (pte_swp_uffd_wp(pte))
1425                         flags |= PM_UFFD_WP;
1426                 entry = pte_to_swp_entry(pte);
1427                 if (pm->show_pfn) {
1428                         pgoff_t offset;
1429                         /*
1430                          * For PFN swap offsets, keeping the offset field
1431                          * to be PFN only to be compatible with old smaps.
1432                          */
1433                         if (is_pfn_swap_entry(entry))
1434                                 offset = swp_offset_pfn(entry);
1435                         else
1436                                 offset = swp_offset(entry);
1437                         frame = swp_type(entry) |
1438                             (offset << MAX_SWAPFILES_SHIFT);
1439                 }
1440                 flags |= PM_SWAP;
1441                 migration = is_migration_entry(entry);
1442                 if (is_pfn_swap_entry(entry))
1443                         page = pfn_swap_entry_to_page(entry);
1444                 if (pte_marker_entry_uffd_wp(entry))
1445                         flags |= PM_UFFD_WP;
1446         }
1447
1448         if (page && !PageAnon(page))
1449                 flags |= PM_FILE;
1450         if (page && !migration && page_mapcount(page) == 1)
1451                 flags |= PM_MMAP_EXCLUSIVE;
1452         if (vma->vm_flags & VM_SOFTDIRTY)
1453                 flags |= PM_SOFT_DIRTY;
1454
1455         return make_pme(frame, flags);
1456 }
1457
1458 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1459                              struct mm_walk *walk)
1460 {
1461         struct vm_area_struct *vma = walk->vma;
1462         struct pagemapread *pm = walk->private;
1463         spinlock_t *ptl;
1464         pte_t *pte, *orig_pte;
1465         int err = 0;
1466 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1467         bool migration = false;
1468
1469         ptl = pmd_trans_huge_lock(pmdp, vma);
1470         if (ptl) {
1471                 u64 flags = 0, frame = 0;
1472                 pmd_t pmd = *pmdp;
1473                 struct page *page = NULL;
1474
1475                 if (vma->vm_flags & VM_SOFTDIRTY)
1476                         flags |= PM_SOFT_DIRTY;
1477
1478                 if (pmd_present(pmd)) {
1479                         page = pmd_page(pmd);
1480
1481                         flags |= PM_PRESENT;
1482                         if (pmd_soft_dirty(pmd))
1483                                 flags |= PM_SOFT_DIRTY;
1484                         if (pmd_uffd_wp(pmd))
1485                                 flags |= PM_UFFD_WP;
1486                         if (pm->show_pfn)
1487                                 frame = pmd_pfn(pmd) +
1488                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1489                 }
1490 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1491                 else if (is_swap_pmd(pmd)) {
1492                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1493                         unsigned long offset;
1494
1495                         if (pm->show_pfn) {
1496                                 if (is_pfn_swap_entry(entry))
1497                                         offset = swp_offset_pfn(entry);
1498                                 else
1499                                         offset = swp_offset(entry);
1500                                 offset = offset +
1501                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1502                                 frame = swp_type(entry) |
1503                                         (offset << MAX_SWAPFILES_SHIFT);
1504                         }
1505                         flags |= PM_SWAP;
1506                         if (pmd_swp_soft_dirty(pmd))
1507                                 flags |= PM_SOFT_DIRTY;
1508                         if (pmd_swp_uffd_wp(pmd))
1509                                 flags |= PM_UFFD_WP;
1510                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1511                         migration = is_migration_entry(entry);
1512                         page = pfn_swap_entry_to_page(entry);
1513                 }
1514 #endif
1515
1516                 if (page && !migration && page_mapcount(page) == 1)
1517                         flags |= PM_MMAP_EXCLUSIVE;
1518
1519                 for (; addr != end; addr += PAGE_SIZE) {
1520                         pagemap_entry_t pme = make_pme(frame, flags);
1521
1522                         err = add_to_pagemap(addr, &pme, pm);
1523                         if (err)
1524                                 break;
1525                         if (pm->show_pfn) {
1526                                 if (flags & PM_PRESENT)
1527                                         frame++;
1528                                 else if (flags & PM_SWAP)
1529                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1530                         }
1531                 }
1532                 spin_unlock(ptl);
1533                 return err;
1534         }
1535
1536         if (pmd_trans_unstable(pmdp))
1537                 return 0;
1538 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1539
1540         /*
1541          * We can assume that @vma always points to a valid one and @end never
1542          * goes beyond vma->vm_end.
1543          */
1544         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1545         for (; addr < end; pte++, addr += PAGE_SIZE) {
1546                 pagemap_entry_t pme;
1547
1548                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1549                 err = add_to_pagemap(addr, &pme, pm);
1550                 if (err)
1551                         break;
1552         }
1553         pte_unmap_unlock(orig_pte, ptl);
1554
1555         cond_resched();
1556
1557         return err;
1558 }
1559
1560 #ifdef CONFIG_HUGETLB_PAGE
1561 /* This function walks within one hugetlb entry in the single call */
1562 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1563                                  unsigned long addr, unsigned long end,
1564                                  struct mm_walk *walk)
1565 {
1566         struct pagemapread *pm = walk->private;
1567         struct vm_area_struct *vma = walk->vma;
1568         u64 flags = 0, frame = 0;
1569         int err = 0;
1570         pte_t pte;
1571
1572         if (vma->vm_flags & VM_SOFTDIRTY)
1573                 flags |= PM_SOFT_DIRTY;
1574
1575         pte = huge_ptep_get(ptep);
1576         if (pte_present(pte)) {
1577                 struct page *page = pte_page(pte);
1578
1579                 if (!PageAnon(page))
1580                         flags |= PM_FILE;
1581
1582                 if (page_mapcount(page) == 1)
1583                         flags |= PM_MMAP_EXCLUSIVE;
1584
1585                 if (huge_pte_uffd_wp(pte))
1586                         flags |= PM_UFFD_WP;
1587
1588                 flags |= PM_PRESENT;
1589                 if (pm->show_pfn)
1590                         frame = pte_pfn(pte) +
1591                                 ((addr & ~hmask) >> PAGE_SHIFT);
1592         } else if (pte_swp_uffd_wp_any(pte)) {
1593                 flags |= PM_UFFD_WP;
1594         }
1595
1596         for (; addr != end; addr += PAGE_SIZE) {
1597                 pagemap_entry_t pme = make_pme(frame, flags);
1598
1599                 err = add_to_pagemap(addr, &pme, pm);
1600                 if (err)
1601                         return err;
1602                 if (pm->show_pfn && (flags & PM_PRESENT))
1603                         frame++;
1604         }
1605
1606         cond_resched();
1607
1608         return err;
1609 }
1610 #else
1611 #define pagemap_hugetlb_range   NULL
1612 #endif /* HUGETLB_PAGE */
1613
1614 static const struct mm_walk_ops pagemap_ops = {
1615         .pmd_entry      = pagemap_pmd_range,
1616         .pte_hole       = pagemap_pte_hole,
1617         .hugetlb_entry  = pagemap_hugetlb_range,
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                 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
1687
1688         /* Ensure the address is inside the task */
1689         if (start_vaddr > mm->task_size)
1690                 start_vaddr = end_vaddr;
1691
1692         /*
1693          * The odds are that this will stop walking way
1694          * before end_vaddr, because the length of the
1695          * user buffer is tracked in "pm", and the walk
1696          * will stop when we hit the end of the buffer.
1697          */
1698         ret = 0;
1699         while (count && (start_vaddr < end_vaddr)) {
1700                 int len;
1701                 unsigned long end;
1702
1703                 pm.pos = 0;
1704                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1705                 /* overflow ? */
1706                 if (end < start_vaddr || end > end_vaddr)
1707                         end = end_vaddr;
1708                 ret = mmap_read_lock_killable(mm);
1709                 if (ret)
1710                         goto out_free;
1711                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1712                 mmap_read_unlock(mm);
1713                 start_vaddr = end;
1714
1715                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1716                 if (copy_to_user(buf, pm.buffer, len)) {
1717                         ret = -EFAULT;
1718                         goto out_free;
1719                 }
1720                 copied += len;
1721                 buf += len;
1722                 count -= len;
1723         }
1724         *ppos += copied;
1725         if (!ret || ret == PM_END_OF_BUFFER)
1726                 ret = copied;
1727
1728 out_free:
1729         kfree(pm.buffer);
1730 out_mm:
1731         mmput(mm);
1732 out:
1733         return ret;
1734 }
1735
1736 static int pagemap_open(struct inode *inode, struct file *file)
1737 {
1738         struct mm_struct *mm;
1739
1740         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1741         if (IS_ERR(mm))
1742                 return PTR_ERR(mm);
1743         file->private_data = mm;
1744         return 0;
1745 }
1746
1747 static int pagemap_release(struct inode *inode, struct file *file)
1748 {
1749         struct mm_struct *mm = file->private_data;
1750
1751         if (mm)
1752                 mmdrop(mm);
1753         return 0;
1754 }
1755
1756 const struct file_operations proc_pagemap_operations = {
1757         .llseek         = mem_lseek, /* borrow this */
1758         .read           = pagemap_read,
1759         .open           = pagemap_open,
1760         .release        = pagemap_release,
1761 };
1762 #endif /* CONFIG_PROC_PAGE_MONITOR */
1763
1764 #ifdef CONFIG_NUMA
1765
1766 struct numa_maps {
1767         unsigned long pages;
1768         unsigned long anon;
1769         unsigned long active;
1770         unsigned long writeback;
1771         unsigned long mapcount_max;
1772         unsigned long dirty;
1773         unsigned long swapcache;
1774         unsigned long node[MAX_NUMNODES];
1775 };
1776
1777 struct numa_maps_private {
1778         struct proc_maps_private proc_maps;
1779         struct numa_maps md;
1780 };
1781
1782 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1783                         unsigned long nr_pages)
1784 {
1785         int count = page_mapcount(page);
1786
1787         md->pages += nr_pages;
1788         if (pte_dirty || PageDirty(page))
1789                 md->dirty += nr_pages;
1790
1791         if (PageSwapCache(page))
1792                 md->swapcache += nr_pages;
1793
1794         if (PageActive(page) || PageUnevictable(page))
1795                 md->active += nr_pages;
1796
1797         if (PageWriteback(page))
1798                 md->writeback += nr_pages;
1799
1800         if (PageAnon(page))
1801                 md->anon += nr_pages;
1802
1803         if (count > md->mapcount_max)
1804                 md->mapcount_max = count;
1805
1806         md->node[page_to_nid(page)] += nr_pages;
1807 }
1808
1809 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1810                 unsigned long addr)
1811 {
1812         struct page *page;
1813         int nid;
1814
1815         if (!pte_present(pte))
1816                 return NULL;
1817
1818         page = vm_normal_page(vma, addr, pte);
1819         if (!page || is_zone_device_page(page))
1820                 return NULL;
1821
1822         if (PageReserved(page))
1823                 return NULL;
1824
1825         nid = page_to_nid(page);
1826         if (!node_isset(nid, node_states[N_MEMORY]))
1827                 return NULL;
1828
1829         return page;
1830 }
1831
1832 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1833 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1834                                               struct vm_area_struct *vma,
1835                                               unsigned long addr)
1836 {
1837         struct page *page;
1838         int nid;
1839
1840         if (!pmd_present(pmd))
1841                 return NULL;
1842
1843         page = vm_normal_page_pmd(vma, addr, pmd);
1844         if (!page)
1845                 return NULL;
1846
1847         if (PageReserved(page))
1848                 return NULL;
1849
1850         nid = page_to_nid(page);
1851         if (!node_isset(nid, node_states[N_MEMORY]))
1852                 return NULL;
1853
1854         return page;
1855 }
1856 #endif
1857
1858 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1859                 unsigned long end, struct mm_walk *walk)
1860 {
1861         struct numa_maps *md = walk->private;
1862         struct vm_area_struct *vma = walk->vma;
1863         spinlock_t *ptl;
1864         pte_t *orig_pte;
1865         pte_t *pte;
1866
1867 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1868         ptl = pmd_trans_huge_lock(pmd, vma);
1869         if (ptl) {
1870                 struct page *page;
1871
1872                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1873                 if (page)
1874                         gather_stats(page, md, pmd_dirty(*pmd),
1875                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1876                 spin_unlock(ptl);
1877                 return 0;
1878         }
1879
1880         if (pmd_trans_unstable(pmd))
1881                 return 0;
1882 #endif
1883         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1884         do {
1885                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1886                 if (!page)
1887                         continue;
1888                 gather_stats(page, md, pte_dirty(*pte), 1);
1889
1890         } while (pte++, addr += PAGE_SIZE, addr != end);
1891         pte_unmap_unlock(orig_pte, ptl);
1892         cond_resched();
1893         return 0;
1894 }
1895 #ifdef CONFIG_HUGETLB_PAGE
1896 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1897                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1898 {
1899         pte_t huge_pte = huge_ptep_get(pte);
1900         struct numa_maps *md;
1901         struct page *page;
1902
1903         if (!pte_present(huge_pte))
1904                 return 0;
1905
1906         page = pte_page(huge_pte);
1907
1908         md = walk->private;
1909         gather_stats(page, md, pte_dirty(huge_pte), 1);
1910         return 0;
1911 }
1912
1913 #else
1914 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1915                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1916 {
1917         return 0;
1918 }
1919 #endif
1920
1921 static const struct mm_walk_ops show_numa_ops = {
1922         .hugetlb_entry = gather_hugetlb_stats,
1923         .pmd_entry = gather_pte_stats,
1924 };
1925
1926 /*
1927  * Display pages allocated per node and memory policy via /proc.
1928  */
1929 static int show_numa_map(struct seq_file *m, void *v)
1930 {
1931         struct numa_maps_private *numa_priv = m->private;
1932         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1933         struct vm_area_struct *vma = v;
1934         struct numa_maps *md = &numa_priv->md;
1935         struct file *file = vma->vm_file;
1936         struct mm_struct *mm = vma->vm_mm;
1937         struct mempolicy *pol;
1938         char buffer[64];
1939         int nid;
1940
1941         if (!mm)
1942                 return 0;
1943
1944         /* Ensure we start with an empty set of numa_maps statistics. */
1945         memset(md, 0, sizeof(*md));
1946
1947         pol = __get_vma_policy(vma, vma->vm_start);
1948         if (pol) {
1949                 mpol_to_str(buffer, sizeof(buffer), pol);
1950                 mpol_cond_put(pol);
1951         } else {
1952                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1953         }
1954
1955         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1956
1957         if (file) {
1958                 seq_puts(m, " file=");
1959                 seq_file_path(m, file, "\n\t= ");
1960         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1961                 seq_puts(m, " heap");
1962         } else if (is_stack(vma)) {
1963                 seq_puts(m, " stack");
1964         }
1965
1966         if (is_vm_hugetlb_page(vma))
1967                 seq_puts(m, " huge");
1968
1969         /* mmap_lock is held by m_start */
1970         walk_page_vma(vma, &show_numa_ops, md);
1971
1972         if (!md->pages)
1973                 goto out;
1974
1975         if (md->anon)
1976                 seq_printf(m, " anon=%lu", md->anon);
1977
1978         if (md->dirty)
1979                 seq_printf(m, " dirty=%lu", md->dirty);
1980
1981         if (md->pages != md->anon && md->pages != md->dirty)
1982                 seq_printf(m, " mapped=%lu", md->pages);
1983
1984         if (md->mapcount_max > 1)
1985                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1986
1987         if (md->swapcache)
1988                 seq_printf(m, " swapcache=%lu", md->swapcache);
1989
1990         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1991                 seq_printf(m, " active=%lu", md->active);
1992
1993         if (md->writeback)
1994                 seq_printf(m, " writeback=%lu", md->writeback);
1995
1996         for_each_node_state(nid, N_MEMORY)
1997                 if (md->node[nid])
1998                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1999
2000         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2001 out:
2002         seq_putc(m, '\n');
2003         return 0;
2004 }
2005
2006 static const struct seq_operations proc_pid_numa_maps_op = {
2007         .start  = m_start,
2008         .next   = m_next,
2009         .stop   = m_stop,
2010         .show   = show_numa_map,
2011 };
2012
2013 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2014 {
2015         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2016                                 sizeof(struct numa_maps_private));
2017 }
2018
2019 const struct file_operations proc_pid_numa_maps_operations = {
2020         .open           = pid_numa_maps_open,
2021         .read           = seq_read,
2022         .llseek         = seq_lseek,
2023         .release        = proc_map_release,
2024 };
2025
2026 #endif /* CONFIG_NUMA */