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                 int mapcount = page_mapcount(page);
 741
 742                 if (mapcount >= 2)
 743                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 744                 else
 745                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 746         }
 747         return 0;
 748 }
 749 #else
 750 #define smaps_hugetlb_range     NULL
 751 #endif /* HUGETLB_PAGE */
 752
 753 static const struct mm_walk_ops smaps_walk_ops = {
 754         .pmd_entry              = smaps_pte_range,
 755         .hugetlb_entry          = smaps_hugetlb_range,
 756 };
 757
 758 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 759         .pmd_entry              = smaps_pte_range,
 760         .hugetlb_entry          = smaps_hugetlb_range,
 761         .pte_hole               = smaps_pte_hole,
 762 };
 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 #ifdef CONFIG_SHMEM
 780         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 781                 /*
 782                  * For shared or readonly shmem mappings we know that all
 783                  * swapped out pages belong to the shmem object, and we can
 784                  * obtain the swap value much more efficiently. For private
 785                  * writable mappings, we might have COW pages that are
 786                  * not affected by the parent swapped out pages of the shmem
 787                  * object, so we have to distinguish them during the page walk.
 788                  * Unless we know that the shmem object (or the part mapped by
 789                  * our VMA) has no swapped out pages at all.
 790                  */
 791                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 792
 793                 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 794                                         !(vma->vm_flags & VM_WRITE))) {
 795                         mss->swap += shmem_swapped;
 796                 } else {
 797                         ops = &smaps_shmem_walk_ops;
 798                 }
 799         }
 800 #endif
 801         /* mmap_lock is held in m_start */
 802         if (!start)
 803                 walk_page_vma(vma, ops, mss);
 804         else
 805                 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
 806 }
 807
 808 #define SEQ_PUT_DEC(str, val) \
 809                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 810
 811 /* Show the contents common for smaps and smaps_rollup */
 812 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 813         bool rollup_mode)
 814 {
 815         SEQ_PUT_DEC("Rss:            ", mss->resident);
 816         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 817         SEQ_PUT_DEC(" kB\nPss_Dirty:      ", mss->pss_dirty >> PSS_SHIFT);
 818         if (rollup_mode) {
 819                 /*
 820                  * These are meaningful only for smaps_rollup, otherwise two of
 821                  * them are zero, and the other one is the same as Pss.
 822                  */
 823                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 824                         mss->pss_anon >> PSS_SHIFT);
 825                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 826                         mss->pss_file >> PSS_SHIFT);
 827                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 828                         mss->pss_shmem >> PSS_SHIFT);
 829         }
 830         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 831         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 832         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 833         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 834         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 835         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 836         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 837         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 838         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 839         SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
 840         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 841         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 842                                   mss->private_hugetlb >> 10, 7);
 843         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 844         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 845                                         mss->swap_pss >> PSS_SHIFT);
 846         SEQ_PUT_DEC(" kB\nLocked:         ",
 847                                         mss->pss_locked >> PSS_SHIFT);
 848         seq_puts(m, " kB\n");
 849 }
 850
 851 static int show_smap(struct seq_file *m, void *v)
 852 {
 853         struct vm_area_struct *vma = v;
 854         struct mem_size_stats mss;
 855
 856         memset(&mss, 0, sizeof(mss));
 857
 858         smap_gather_stats(vma, &mss, 0);
 859
 860         show_map_vma(m, vma);
 861
 862         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 863         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 864         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 865         seq_puts(m, " kB\n");
 866
 867         __show_smap(m, &mss, false);
 868
 869         seq_printf(m, "THPeligible:    %d\n",
 870                    hugepage_vma_check(vma, vma->vm_flags, true, false, true));
 871
 872         if (arch_pkeys_enabled())
 873                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 874         show_smap_vma_flags(m, vma);
 875
 876         return 0;
 877 }
 878
 879 static int show_smaps_rollup(struct seq_file *m, void *v)
 880 {
 881         struct proc_maps_private *priv = m->private;
 882         struct mem_size_stats mss;
 883         struct mm_struct *mm = priv->mm;
 884         struct vm_area_struct *vma;
 885         unsigned long vma_start = 0, last_vma_end = 0;
 886         int ret = 0;
 887         MA_STATE(mas, &mm->mm_mt, 0, 0);
 888
 889         priv->task = get_proc_task(priv->inode);
 890         if (!priv->task)
 891                 return -ESRCH;
 892
 893         if (!mm || !mmget_not_zero(mm)) {
 894                 ret = -ESRCH;
 895                 goto out_put_task;
 896         }
 897
 898         memset(&mss, 0, sizeof(mss));
 899
 900         ret = mmap_read_lock_killable(mm);
 901         if (ret)
 902                 goto out_put_mm;
 903
 904         hold_task_mempolicy(priv);
 905         vma = mas_find(&mas, ULONG_MAX);
 906
 907         if (unlikely(!vma))
 908                 goto empty_set;
 909
 910         vma_start = vma->vm_start;
 911         do {
 912                 smap_gather_stats(vma, &mss, 0);
 913                 last_vma_end = vma->vm_end;
 914
 915                 /*
 916                  * Release mmap_lock temporarily if someone wants to
 917                  * access it for write request.
 918                  */
 919                 if (mmap_lock_is_contended(mm)) {
 920                         mas_pause(&mas);
 921                         mmap_read_unlock(mm);
 922                         ret = mmap_read_lock_killable(mm);
 923                         if (ret) {
 924                                 release_task_mempolicy(priv);
 925                                 goto out_put_mm;
 926                         }
 927
 928                         /*
 929                          * After dropping the lock, there are four cases to
 930                          * consider. See the following example for explanation.
 931                          *
 932                          *   +------+------+-----------+
 933                          *   | VMA1 | VMA2 | VMA3      |
 934                          *   +------+------+-----------+
 935                          *   |      |      |           |
 936                          *  4k     8k     16k         400k
 937                          *
 938                          * Suppose we drop the lock after reading VMA2 due to
 939                          * contention, then we get:
 940                          *
 941                          *      last_vma_end = 16k
 942                          *
 943                          * 1) VMA2 is freed, but VMA3 exists:
 944                          *
 945                          *    find_vma(mm, 16k - 1) will return VMA3.
 946                          *    In this case, just continue from VMA3.
 947                          *
 948                          * 2) VMA2 still exists:
 949                          *
 950                          *    find_vma(mm, 16k - 1) will return VMA2.
 951                          *    Iterate the loop like the original one.
 952                          *
 953                          * 3) No more VMAs can be found:
 954                          *
 955                          *    find_vma(mm, 16k - 1) will return NULL.
 956                          *    No more things to do, just break.
 957                          *
 958                          * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
 959                          *
 960                          *    find_vma(mm, 16k - 1) will return VMA' whose range
 961                          *    contains last_vma_end.
 962                          *    Iterate VMA' from last_vma_end.
 963                          */
 964                         vma = mas_find(&mas, ULONG_MAX);
 965                         /* Case 3 above */
 966                         if (!vma)
 967                                 break;
 968
 969                         /* Case 1 above */
 970                         if (vma->vm_start >= last_vma_end)
 971                                 continue;
 972
 973                         /* Case 4 above */
 974                         if (vma->vm_end > last_vma_end)
 975                                 smap_gather_stats(vma, &mss, last_vma_end);
 976                 }
 977                 /* Case 2 above */
 978         } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL);
 979
 980 empty_set:
 981         show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
 982         seq_pad(m, ' ');
 983         seq_puts(m, "[rollup]\n");
 984
 985         __show_smap(m, &mss, true);
 986
 987         release_task_mempolicy(priv);
 988         mmap_read_unlock(mm);
 989
 990 out_put_mm:
 991         mmput(mm);
 992 out_put_task:
 993         put_task_struct(priv->task);
 994         priv->task = NULL;
 995
 996         return ret;
 997 }
 998 #undef SEQ_PUT_DEC
 999
1000 static const struct seq_operations proc_pid_smaps_op = {
1001         .start  = m_start,
1002         .next   = m_next,
1003         .stop   = m_stop,
1004         .show   = show_smap
1005 };
1006
1007 static int pid_smaps_open(struct inode *inode, struct file *file)
1008 {
1009         return do_maps_open(inode, file, &proc_pid_smaps_op);
1010 }
1011
1012 static int smaps_rollup_open(struct inode *inode, struct file *file)
1013 {
1014         int ret;
1015         struct proc_maps_private *priv;
1016
1017         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1018         if (!priv)
1019                 return -ENOMEM;
1020
1021         ret = single_open(file, show_smaps_rollup, priv);
1022         if (ret)
1023                 goto out_free;
1024
1025         priv->inode = inode;
1026         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1027         if (IS_ERR(priv->mm)) {
1028                 ret = PTR_ERR(priv->mm);
1029
1030                 single_release(inode, file);
1031                 goto out_free;
1032         }
1033
1034         return 0;
1035
1036 out_free:
1037         kfree(priv);
1038         return ret;
1039 }
1040
1041 static int smaps_rollup_release(struct inode *inode, struct file *file)
1042 {
1043         struct seq_file *seq = file->private_data;
1044         struct proc_maps_private *priv = seq->private;
1045
1046         if (priv->mm)
1047                 mmdrop(priv->mm);
1048
1049         kfree(priv);
1050         return single_release(inode, file);
1051 }
1052
1053 const struct file_operations proc_pid_smaps_operations = {
1054         .open           = pid_smaps_open,
1055         .read           = seq_read,
1056         .llseek         = seq_lseek,
1057         .release        = proc_map_release,
1058 };
1059
1060 const struct file_operations proc_pid_smaps_rollup_operations = {
1061         .open           = smaps_rollup_open,
1062         .read           = seq_read,
1063         .llseek         = seq_lseek,
1064         .release        = smaps_rollup_release,
1065 };
1066
1067 enum clear_refs_types {
1068         CLEAR_REFS_ALL = 1,
1069         CLEAR_REFS_ANON,
1070         CLEAR_REFS_MAPPED,
1071         CLEAR_REFS_SOFT_DIRTY,
1072         CLEAR_REFS_MM_HIWATER_RSS,
1073         CLEAR_REFS_LAST,
1074 };
1075
1076 struct clear_refs_private {
1077         enum clear_refs_types type;
1078 };
1079
1080 #ifdef CONFIG_MEM_SOFT_DIRTY
1081
1082 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1083 {
1084         struct page *page;
1085
1086         if (!pte_write(pte))
1087                 return false;
1088         if (!is_cow_mapping(vma->vm_flags))
1089                 return false;
1090         if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1091                 return false;
1092         page = vm_normal_page(vma, addr, pte);
1093         if (!page)
1094                 return false;
1095         return page_maybe_dma_pinned(page);
1096 }
1097
1098 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1099                 unsigned long addr, pte_t *pte)
1100 {
1101         /*
1102          * The soft-dirty tracker uses #PF-s to catch writes
1103          * to pages, so write-protect the pte as well. See the
1104          * Documentation/admin-guide/mm/soft-dirty.rst for full description
1105          * of how soft-dirty works.
1106          */
1107         pte_t ptent = *pte;
1108
1109         if (pte_present(ptent)) {
1110                 pte_t old_pte;
1111
1112                 if (pte_is_pinned(vma, addr, ptent))
1113                         return;
1114                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1115                 ptent = pte_wrprotect(old_pte);
1116                 ptent = pte_clear_soft_dirty(ptent);
1117                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1118         } else if (is_swap_pte(ptent)) {
1119                 ptent = pte_swp_clear_soft_dirty(ptent);
1120                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1121         }
1122 }
1123 #else
1124 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1125                 unsigned long addr, pte_t *pte)
1126 {
1127 }
1128 #endif
1129
1130 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1131 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1132                 unsigned long addr, pmd_t *pmdp)
1133 {
1134         pmd_t old, pmd = *pmdp;
1135
1136         if (pmd_present(pmd)) {
1137                 /* See comment in change_huge_pmd() */
1138                 old = pmdp_invalidate(vma, addr, pmdp);
1139                 if (pmd_dirty(old))
1140                         pmd = pmd_mkdirty(pmd);
1141                 if (pmd_young(old))
1142                         pmd = pmd_mkyoung(pmd);
1143
1144                 pmd = pmd_wrprotect(pmd);
1145                 pmd = pmd_clear_soft_dirty(pmd);
1146
1147                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1148         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1149                 pmd = pmd_swp_clear_soft_dirty(pmd);
1150                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1151         }
1152 }
1153 #else
1154 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1155                 unsigned long addr, pmd_t *pmdp)
1156 {
1157 }
1158 #endif
1159
1160 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1161                                 unsigned long end, struct mm_walk *walk)
1162 {
1163         struct clear_refs_private *cp = walk->private;
1164         struct vm_area_struct *vma = walk->vma;
1165         pte_t *pte, ptent;
1166         spinlock_t *ptl;
1167         struct page *page;
1168
1169         ptl = pmd_trans_huge_lock(pmd, vma);
1170         if (ptl) {
1171                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1172                         clear_soft_dirty_pmd(vma, addr, pmd);
1173                         goto out;
1174                 }
1175
1176                 if (!pmd_present(*pmd))
1177                         goto out;
1178
1179                 page = pmd_page(*pmd);
1180
1181                 /* Clear accessed and referenced bits. */
1182                 pmdp_test_and_clear_young(vma, addr, pmd);
1183                 test_and_clear_page_young(page);
1184                 ClearPageReferenced(page);
1185 out:
1186                 spin_unlock(ptl);
1187                 return 0;
1188         }
1189
1190         if (pmd_trans_unstable(pmd))
1191                 return 0;
1192
1193         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1194         for (; addr != end; pte++, addr += PAGE_SIZE) {
1195                 ptent = *pte;
1196
1197                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1198                         clear_soft_dirty(vma, addr, pte);
1199                         continue;
1200                 }
1201
1202                 if (!pte_present(ptent))
1203                         continue;
1204
1205                 page = vm_normal_page(vma, addr, ptent);
1206                 if (!page)
1207                         continue;
1208
1209                 /* Clear accessed and referenced bits. */
1210                 ptep_test_and_clear_young(vma, addr, pte);
1211                 test_and_clear_page_young(page);
1212                 ClearPageReferenced(page);
1213         }
1214         pte_unmap_unlock(pte - 1, ptl);
1215         cond_resched();
1216         return 0;
1217 }
1218
1219 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1220                                 struct mm_walk *walk)
1221 {
1222         struct clear_refs_private *cp = walk->private;
1223         struct vm_area_struct *vma = walk->vma;
1224
1225         if (vma->vm_flags & VM_PFNMAP)
1226                 return 1;
1227
1228         /*
1229          * Writing 1 to /proc/pid/clear_refs affects all pages.
1230          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1231          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1232          * Writing 4 to /proc/pid/clear_refs affects all pages.
1233          */
1234         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1235                 return 1;
1236         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1237                 return 1;
1238         return 0;
1239 }
1240
1241 static const struct mm_walk_ops clear_refs_walk_ops = {
1242         .pmd_entry              = clear_refs_pte_range,
1243         .test_walk              = clear_refs_test_walk,
1244 };
1245
1246 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1247                                 size_t count, loff_t *ppos)
1248 {
1249         struct task_struct *task;
1250         char buffer[PROC_NUMBUF];
1251         struct mm_struct *mm;
1252         struct vm_area_struct *vma;
1253         enum clear_refs_types type;
1254         int itype;
1255         int rv;
1256
1257         memset(buffer, 0, sizeof(buffer));
1258         if (count > sizeof(buffer) - 1)
1259                 count = sizeof(buffer) - 1;
1260         if (copy_from_user(buffer, buf, count))
1261                 return -EFAULT;
1262         rv = kstrtoint(strstrip(buffer), 10, &itype);
1263         if (rv < 0)
1264                 return rv;
1265         type = (enum clear_refs_types)itype;
1266         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1267                 return -EINVAL;
1268
1269         task = get_proc_task(file_inode(file));
1270         if (!task)
1271                 return -ESRCH;
1272         mm = get_task_mm(task);
1273         if (mm) {
1274                 MA_STATE(mas, &mm->mm_mt, 0, 0);
1275                 struct mmu_notifier_range range;
1276                 struct clear_refs_private cp = {
1277                         .type = type,
1278                 };
1279
1280                 if (mmap_write_lock_killable(mm)) {
1281                         count = -EINTR;
1282                         goto out_mm;
1283                 }
1284                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1285                         /*
1286                          * Writing 5 to /proc/pid/clear_refs resets the peak
1287                          * resident set size to this mm's current rss value.
1288                          */
1289                         reset_mm_hiwater_rss(mm);
1290                         goto out_unlock;
1291                 }
1292
1293                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1294                         mas_for_each(&mas, vma, ULONG_MAX) {
1295                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1296                                         continue;
1297                                 vma->vm_flags &= ~VM_SOFTDIRTY;
1298                                 vma_set_page_prot(vma);
1299                         }
1300
1301                         inc_tlb_flush_pending(mm);
1302                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1303                                                 0, NULL, mm, 0, -1UL);
1304                         mmu_notifier_invalidate_range_start(&range);
1305                 }
1306                 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1307                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1308                         mmu_notifier_invalidate_range_end(&range);
1309                         flush_tlb_mm(mm);
1310                         dec_tlb_flush_pending(mm);
1311                 }
1312 out_unlock:
1313                 mmap_write_unlock(mm);
1314 out_mm:
1315                 mmput(mm);
1316         }
1317         put_task_struct(task);
1318
1319         return count;
1320 }
1321
1322 const struct file_operations proc_clear_refs_operations = {
1323         .write          = clear_refs_write,
1324         .llseek         = noop_llseek,
1325 };
1326
1327 typedef struct {
1328         u64 pme;
1329 } pagemap_entry_t;
1330
1331 struct pagemapread {
1332         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1333         pagemap_entry_t *buffer;
1334         bool show_pfn;
1335 };
1336
1337 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1338 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1339
1340 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1341 #define PM_PFRAME_BITS          55
1342 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1343 #define PM_SOFT_DIRTY           BIT_ULL(55)
1344 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1345 #define PM_UFFD_WP              BIT_ULL(57)
1346 #define PM_FILE                 BIT_ULL(61)
1347 #define PM_SWAP                 BIT_ULL(62)
1348 #define PM_PRESENT              BIT_ULL(63)
1349
1350 #define PM_END_OF_BUFFER    1
1351
1352 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1353 {
1354         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1355 }
1356
1357 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1358                           struct pagemapread *pm)
1359 {
1360         pm->buffer[pm->pos++] = *pme;
1361         if (pm->pos >= pm->len)
1362                 return PM_END_OF_BUFFER;
1363         return 0;
1364 }
1365
1366 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1367                             __always_unused int depth, struct mm_walk *walk)
1368 {
1369         struct pagemapread *pm = walk->private;
1370         unsigned long addr = start;
1371         int err = 0;
1372
1373         while (addr < end) {
1374                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1375                 pagemap_entry_t pme = make_pme(0, 0);
1376                 /* End of address space hole, which we mark as non-present. */
1377                 unsigned long hole_end;
1378
1379                 if (vma)
1380                         hole_end = min(end, vma->vm_start);
1381                 else
1382                         hole_end = end;
1383
1384                 for (; addr < hole_end; addr += PAGE_SIZE) {
1385                         err = add_to_pagemap(addr, &pme, pm);
1386                         if (err)
1387                                 goto out;
1388                 }
1389
1390                 if (!vma)
1391                         break;
1392
1393                 /* Addresses in the VMA. */
1394                 if (vma->vm_flags & VM_SOFTDIRTY)
1395                         pme = make_pme(0, PM_SOFT_DIRTY);
1396                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1397                         err = add_to_pagemap(addr, &pme, pm);
1398                         if (err)
1399                                 goto out;
1400                 }
1401         }
1402 out:
1403         return err;
1404 }
1405
1406 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1407                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1408 {
1409         u64 frame = 0, flags = 0;
1410         struct page *page = NULL;
1411         bool migration = false;
1412
1413         if (pte_present(pte)) {
1414                 if (pm->show_pfn)
1415                         frame = pte_pfn(pte);
1416                 flags |= PM_PRESENT;
1417                 page = vm_normal_page(vma, addr, pte);
1418                 if (pte_soft_dirty(pte))
1419                         flags |= PM_SOFT_DIRTY;
1420                 if (pte_uffd_wp(pte))
1421                         flags |= PM_UFFD_WP;
1422         } else if (is_swap_pte(pte)) {
1423                 swp_entry_t entry;
1424                 if (pte_swp_soft_dirty(pte))
1425                         flags |= PM_SOFT_DIRTY;
1426                 if (pte_swp_uffd_wp(pte))
1427                         flags |= PM_UFFD_WP;
1428                 entry = pte_to_swp_entry(pte);
1429                 if (pm->show_pfn) {
1430                         pgoff_t offset;
1431                         /*
1432                          * For PFN swap offsets, keeping the offset field
1433                          * to be PFN only to be compatible with old smaps.
1434                          */
1435                         if (is_pfn_swap_entry(entry))
1436                                 offset = swp_offset_pfn(entry);
1437                         else
1438                                 offset = swp_offset(entry);
1439                         frame = swp_type(entry) |
1440                             (offset << MAX_SWAPFILES_SHIFT);
1441                 }
1442                 flags |= PM_SWAP;
1443                 migration = is_migration_entry(entry);
1444                 if (is_pfn_swap_entry(entry))
1445                         page = pfn_swap_entry_to_page(entry);
1446                 if (pte_marker_entry_uffd_wp(entry))
1447                         flags |= PM_UFFD_WP;
1448         }
1449
1450         if (page && !PageAnon(page))
1451                 flags |= PM_FILE;
1452         if (page && !migration && page_mapcount(page) == 1)
1453                 flags |= PM_MMAP_EXCLUSIVE;
1454         if (vma->vm_flags & VM_SOFTDIRTY)
1455                 flags |= PM_SOFT_DIRTY;
1456
1457         return make_pme(frame, flags);
1458 }
1459
1460 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1461                              struct mm_walk *walk)
1462 {
1463         struct vm_area_struct *vma = walk->vma;
1464         struct pagemapread *pm = walk->private;
1465         spinlock_t *ptl;
1466         pte_t *pte, *orig_pte;
1467         int err = 0;
1468 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1469         bool migration = false;
1470
1471         ptl = pmd_trans_huge_lock(pmdp, vma);
1472         if (ptl) {
1473                 u64 flags = 0, frame = 0;
1474                 pmd_t pmd = *pmdp;
1475                 struct page *page = NULL;
1476
1477                 if (vma->vm_flags & VM_SOFTDIRTY)
1478                         flags |= PM_SOFT_DIRTY;
1479
1480                 if (pmd_present(pmd)) {
1481                         page = pmd_page(pmd);
1482
1483                         flags |= PM_PRESENT;
1484                         if (pmd_soft_dirty(pmd))
1485                                 flags |= PM_SOFT_DIRTY;
1486                         if (pmd_uffd_wp(pmd))
1487                                 flags |= PM_UFFD_WP;
1488                         if (pm->show_pfn)
1489                                 frame = pmd_pfn(pmd) +
1490                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1491                 }
1492 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1493                 else if (is_swap_pmd(pmd)) {
1494                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1495                         unsigned long offset;
1496
1497                         if (pm->show_pfn) {
1498                                 if (is_pfn_swap_entry(entry))
1499                                         offset = swp_offset_pfn(entry);
1500                                 else
1501                                         offset = swp_offset(entry);
1502                                 offset = offset +
1503                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1504                                 frame = swp_type(entry) |
1505                                         (offset << MAX_SWAPFILES_SHIFT);
1506                         }
1507                         flags |= PM_SWAP;
1508                         if (pmd_swp_soft_dirty(pmd))
1509                                 flags |= PM_SOFT_DIRTY;
1510                         if (pmd_swp_uffd_wp(pmd))
1511                                 flags |= PM_UFFD_WP;
1512                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1513                         migration = is_migration_entry(entry);
1514                         page = pfn_swap_entry_to_page(entry);
1515                 }
1516 #endif
1517
1518                 if (page && !migration && page_mapcount(page) == 1)
1519                         flags |= PM_MMAP_EXCLUSIVE;
1520
1521                 for (; addr != end; addr += PAGE_SIZE) {
1522                         pagemap_entry_t pme = make_pme(frame, flags);
1523
1524                         err = add_to_pagemap(addr, &pme, pm);
1525                         if (err)
1526                                 break;
1527                         if (pm->show_pfn) {
1528                                 if (flags & PM_PRESENT)
1529                                         frame++;
1530                                 else if (flags & PM_SWAP)
1531                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1532                         }
1533                 }
1534                 spin_unlock(ptl);
1535                 return err;
1536         }
1537
1538         if (pmd_trans_unstable(pmdp))
1539                 return 0;
1540 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1541
1542         /*
1543          * We can assume that @vma always points to a valid one and @end never
1544          * goes beyond vma->vm_end.
1545          */
1546         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1547         for (; addr < end; pte++, addr += PAGE_SIZE) {
1548                 pagemap_entry_t pme;
1549
1550                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1551                 err = add_to_pagemap(addr, &pme, pm);
1552                 if (err)
1553                         break;
1554         }
1555         pte_unmap_unlock(orig_pte, ptl);
1556
1557         cond_resched();
1558
1559         return err;
1560 }
1561
1562 #ifdef CONFIG_HUGETLB_PAGE
1563 /* This function walks within one hugetlb entry in the single call */
1564 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1565                                  unsigned long addr, unsigned long end,
1566                                  struct mm_walk *walk)
1567 {
1568         struct pagemapread *pm = walk->private;
1569         struct vm_area_struct *vma = walk->vma;
1570         u64 flags = 0, frame = 0;
1571         int err = 0;
1572         pte_t pte;
1573
1574         if (vma->vm_flags & VM_SOFTDIRTY)
1575                 flags |= PM_SOFT_DIRTY;
1576
1577         pte = huge_ptep_get(ptep);
1578         if (pte_present(pte)) {
1579                 struct page *page = pte_page(pte);
1580
1581                 if (!PageAnon(page))
1582                         flags |= PM_FILE;
1583
1584                 if (page_mapcount(page) == 1)
1585                         flags |= PM_MMAP_EXCLUSIVE;
1586
1587                 if (huge_pte_uffd_wp(pte))
1588                         flags |= PM_UFFD_WP;
1589
1590                 flags |= PM_PRESENT;
1591                 if (pm->show_pfn)
1592                         frame = pte_pfn(pte) +
1593                                 ((addr & ~hmask) >> PAGE_SHIFT);
1594         } else if (pte_swp_uffd_wp_any(pte)) {
1595                 flags |= PM_UFFD_WP;
1596         }
1597
1598         for (; addr != end; addr += PAGE_SIZE) {
1599                 pagemap_entry_t pme = make_pme(frame, flags);
1600
1601                 err = add_to_pagemap(addr, &pme, pm);
1602                 if (err)
1603                         return err;
1604                 if (pm->show_pfn && (flags & PM_PRESENT))
1605                         frame++;
1606         }
1607
1608         cond_resched();
1609
1610         return err;
1611 }
1612 #else
1613 #define pagemap_hugetlb_range   NULL
1614 #endif /* HUGETLB_PAGE */
1615
1616 static const struct mm_walk_ops pagemap_ops = {
1617         .pmd_entry      = pagemap_pmd_range,
1618         .pte_hole       = pagemap_pte_hole,
1619         .hugetlb_entry  = pagemap_hugetlb_range,
1620 };
1621
1622 /*
1623  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1624  *
1625  * For each page in the address space, this file contains one 64-bit entry
1626  * consisting of the following:
1627  *
1628  * Bits 0-54  page frame number (PFN) if present
1629  * Bits 0-4   swap type if swapped
1630  * Bits 5-54  swap offset if swapped
1631  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1632  * Bit  56    page exclusively mapped
1633  * Bit  57    pte is uffd-wp write-protected
1634  * Bits 58-60 zero
1635  * Bit  61    page is file-page or shared-anon
1636  * Bit  62    page swapped
1637  * Bit  63    page present
1638  *
1639  * If the page is not present but in swap, then the PFN contains an
1640  * encoding of the swap file number and the page's offset into the
1641  * swap. Unmapped pages return a null PFN. This allows determining
1642  * precisely which pages are mapped (or in swap) and comparing mapped
1643  * pages between processes.
1644  *
1645  * Efficient users of this interface will use /proc/pid/maps to
1646  * determine which areas of memory are actually mapped and llseek to
1647  * skip over unmapped regions.
1648  */
1649 static ssize_t pagemap_read(struct file *file, char __user *buf,
1650                             size_t count, loff_t *ppos)
1651 {
1652         struct mm_struct *mm = file->private_data;
1653         struct pagemapread pm;
1654         unsigned long src;
1655         unsigned long svpfn;
1656         unsigned long start_vaddr;
1657         unsigned long end_vaddr;
1658         int ret = 0, copied = 0;
1659
1660         if (!mm || !mmget_not_zero(mm))
1661                 goto out;
1662
1663         ret = -EINVAL;
1664         /* file position must be aligned */
1665         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1666                 goto out_mm;
1667
1668         ret = 0;
1669         if (!count)
1670                 goto out_mm;
1671
1672         /* do not disclose physical addresses: attack vector */
1673         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1674
1675         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1676         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1677         ret = -ENOMEM;
1678         if (!pm.buffer)
1679                 goto out_mm;
1680
1681         src = *ppos;
1682         svpfn = src / PM_ENTRY_BYTES;
1683         end_vaddr = mm->task_size;
1684
1685         /* watch out for wraparound */
1686         start_vaddr = end_vaddr;
1687         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT))
1688                 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
1689
1690         /* Ensure the address is inside the task */
1691         if (start_vaddr > mm->task_size)
1692                 start_vaddr = end_vaddr;
1693
1694         /*
1695          * The odds are that this will stop walking way
1696          * before end_vaddr, because the length of the
1697          * user buffer is tracked in "pm", and the walk
1698          * will stop when we hit the end of the buffer.
1699          */
1700         ret = 0;
1701         while (count && (start_vaddr < end_vaddr)) {
1702                 int len;
1703                 unsigned long end;
1704
1705                 pm.pos = 0;
1706                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1707                 /* overflow ? */
1708                 if (end < start_vaddr || end > end_vaddr)
1709                         end = end_vaddr;
1710                 ret = mmap_read_lock_killable(mm);
1711                 if (ret)
1712                         goto out_free;
1713                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1714                 mmap_read_unlock(mm);
1715                 start_vaddr = end;
1716
1717                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1718                 if (copy_to_user(buf, pm.buffer, len)) {
1719                         ret = -EFAULT;
1720                         goto out_free;
1721                 }
1722                 copied += len;
1723                 buf += len;
1724                 count -= len;
1725         }
1726         *ppos += copied;
1727         if (!ret || ret == PM_END_OF_BUFFER)
1728                 ret = copied;
1729
1730 out_free:
1731         kfree(pm.buffer);
1732 out_mm:
1733         mmput(mm);
1734 out:
1735         return ret;
1736 }
1737
1738 static int pagemap_open(struct inode *inode, struct file *file)
1739 {
1740         struct mm_struct *mm;
1741
1742         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1743         if (IS_ERR(mm))
1744                 return PTR_ERR(mm);
1745         file->private_data = mm;
1746         return 0;
1747 }
1748
1749 static int pagemap_release(struct inode *inode, struct file *file)
1750 {
1751         struct mm_struct *mm = file->private_data;
1752
1753         if (mm)
1754                 mmdrop(mm);
1755         return 0;
1756 }
1757
1758 const struct file_operations proc_pagemap_operations = {
1759         .llseek         = mem_lseek, /* borrow this */
1760         .read           = pagemap_read,
1761         .open           = pagemap_open,
1762         .release        = pagemap_release,
1763 };
1764 #endif /* CONFIG_PROC_PAGE_MONITOR */
1765
1766 #ifdef CONFIG_NUMA
1767
1768 struct numa_maps {
1769         unsigned long pages;
1770         unsigned long anon;
1771         unsigned long active;
1772         unsigned long writeback;
1773         unsigned long mapcount_max;
1774         unsigned long dirty;
1775         unsigned long swapcache;
1776         unsigned long node[MAX_NUMNODES];
1777 };
1778
1779 struct numa_maps_private {
1780         struct proc_maps_private proc_maps;
1781         struct numa_maps md;
1782 };
1783
1784 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1785                         unsigned long nr_pages)
1786 {
1787         int count = page_mapcount(page);
1788
1789         md->pages += nr_pages;
1790         if (pte_dirty || PageDirty(page))
1791                 md->dirty += nr_pages;
1792
1793         if (PageSwapCache(page))
1794                 md->swapcache += nr_pages;
1795
1796         if (PageActive(page) || PageUnevictable(page))
1797                 md->active += nr_pages;
1798
1799         if (PageWriteback(page))
1800                 md->writeback += nr_pages;
1801
1802         if (PageAnon(page))
1803                 md->anon += nr_pages;
1804
1805         if (count > md->mapcount_max)
1806                 md->mapcount_max = count;
1807
1808         md->node[page_to_nid(page)] += nr_pages;
1809 }
1810
1811 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1812                 unsigned long addr)
1813 {
1814         struct page *page;
1815         int nid;
1816
1817         if (!pte_present(pte))
1818                 return NULL;
1819
1820         page = vm_normal_page(vma, addr, pte);
1821         if (!page || is_zone_device_page(page))
1822                 return NULL;
1823
1824         if (PageReserved(page))
1825                 return NULL;
1826
1827         nid = page_to_nid(page);
1828         if (!node_isset(nid, node_states[N_MEMORY]))
1829                 return NULL;
1830
1831         return page;
1832 }
1833
1834 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1835 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1836                                               struct vm_area_struct *vma,
1837                                               unsigned long addr)
1838 {
1839         struct page *page;
1840         int nid;
1841
1842         if (!pmd_present(pmd))
1843                 return NULL;
1844
1845         page = vm_normal_page_pmd(vma, addr, pmd);
1846         if (!page)
1847                 return NULL;
1848
1849         if (PageReserved(page))
1850                 return NULL;
1851
1852         nid = page_to_nid(page);
1853         if (!node_isset(nid, node_states[N_MEMORY]))
1854                 return NULL;
1855
1856         return page;
1857 }
1858 #endif
1859
1860 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1861                 unsigned long end, struct mm_walk *walk)
1862 {
1863         struct numa_maps *md = walk->private;
1864         struct vm_area_struct *vma = walk->vma;
1865         spinlock_t *ptl;
1866         pte_t *orig_pte;
1867         pte_t *pte;
1868
1869 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1870         ptl = pmd_trans_huge_lock(pmd, vma);
1871         if (ptl) {
1872                 struct page *page;
1873
1874                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1875                 if (page)
1876                         gather_stats(page, md, pmd_dirty(*pmd),
1877                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1878                 spin_unlock(ptl);
1879                 return 0;
1880         }
1881
1882         if (pmd_trans_unstable(pmd))
1883                 return 0;
1884 #endif
1885         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1886         do {
1887                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1888                 if (!page)
1889                         continue;
1890                 gather_stats(page, md, pte_dirty(*pte), 1);
1891
1892         } while (pte++, addr += PAGE_SIZE, addr != end);
1893         pte_unmap_unlock(orig_pte, ptl);
1894         cond_resched();
1895         return 0;
1896 }
1897 #ifdef CONFIG_HUGETLB_PAGE
1898 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1899                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1900 {
1901         pte_t huge_pte = huge_ptep_get(pte);
1902         struct numa_maps *md;
1903         struct page *page;
1904
1905         if (!pte_present(huge_pte))
1906                 return 0;
1907
1908         page = pte_page(huge_pte);
1909
1910         md = walk->private;
1911         gather_stats(page, md, pte_dirty(huge_pte), 1);
1912         return 0;
1913 }
1914
1915 #else
1916 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1917                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1918 {
1919         return 0;
1920 }
1921 #endif
1922
1923 static const struct mm_walk_ops show_numa_ops = {
1924         .hugetlb_entry = gather_hugetlb_stats,
1925         .pmd_entry = gather_pte_stats,
1926 };
1927
1928 /*
1929  * Display pages allocated per node and memory policy via /proc.
1930  */
1931 static int show_numa_map(struct seq_file *m, void *v)
1932 {
1933         struct numa_maps_private *numa_priv = m->private;
1934         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1935         struct vm_area_struct *vma = v;
1936         struct numa_maps *md = &numa_priv->md;
1937         struct file *file = vma->vm_file;
1938         struct mm_struct *mm = vma->vm_mm;
1939         struct mempolicy *pol;
1940         char buffer[64];
1941         int nid;
1942
1943         if (!mm)
1944                 return 0;
1945
1946         /* Ensure we start with an empty set of numa_maps statistics. */
1947         memset(md, 0, sizeof(*md));
1948
1949         pol = __get_vma_policy(vma, vma->vm_start);
1950         if (pol) {
1951                 mpol_to_str(buffer, sizeof(buffer), pol);
1952                 mpol_cond_put(pol);
1953         } else {
1954                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1955         }
1956
1957         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1958
1959         if (file) {
1960                 seq_puts(m, " file=");
1961                 seq_file_path(m, file, "\n\t= ");
1962         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1963                 seq_puts(m, " heap");
1964         } else if (is_stack(vma)) {
1965                 seq_puts(m, " stack");
1966         }
1967
1968         if (is_vm_hugetlb_page(vma))
1969                 seq_puts(m, " huge");
1970
1971         /* mmap_lock is held by m_start */
1972         walk_page_vma(vma, &show_numa_ops, md);
1973
1974         if (!md->pages)
1975                 goto out;
1976
1977         if (md->anon)
1978                 seq_printf(m, " anon=%lu", md->anon);
1979
1980         if (md->dirty)
1981                 seq_printf(m, " dirty=%lu", md->dirty);
1982
1983         if (md->pages != md->anon && md->pages != md->dirty)
1984                 seq_printf(m, " mapped=%lu", md->pages);
1985
1986         if (md->mapcount_max > 1)
1987                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1988
1989         if (md->swapcache)
1990                 seq_printf(m, " swapcache=%lu", md->swapcache);
1991
1992         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1993                 seq_printf(m, " active=%lu", md->active);
1994
1995         if (md->writeback)
1996                 seq_printf(m, " writeback=%lu", md->writeback);
1997
1998         for_each_node_state(nid, N_MEMORY)
1999                 if (md->node[nid])
2000                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2001
2002         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2003 out:
2004         seq_putc(m, '\n');
2005         return 0;
2006 }
2007
2008 static const struct seq_operations proc_pid_numa_maps_op = {
2009         .start  = m_start,
2010         .next   = m_next,
2011         .stop   = m_stop,
2012         .show   = show_numa_map,
2013 };
2014
2015 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2016 {
2017         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2018                                 sizeof(struct numa_maps_private));
2019 }
2020
2021 const struct file_operations proc_pid_numa_maps_operations = {
2022         .open           = pid_numa_maps_open,
2023         .read           = seq_read,
2024         .llseek         = seq_lseek,
2025         .release        = proc_map_release,
2026 };
2027
2028 #endif /* CONFIG_NUMA */