fs/proc/task_mmu.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/pagewalk.h>
   3 #include <linux/vmacache.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 void *m_start(struct seq_file *m, loff_t *ppos)
 127 {
 128         struct proc_maps_private *priv = m->private;
 129         unsigned long last_addr = *ppos;
 130         struct mm_struct *mm;
 131         struct vm_area_struct *vma;
 132
 133         /* See m_next(). Zero at the start or after lseek. */
 134         if (last_addr == -1UL)
 135                 return NULL;
 136
 137         priv->task = get_proc_task(priv->inode);
 138         if (!priv->task)
 139                 return ERR_PTR(-ESRCH);
 140
 141         mm = priv->mm;
 142         if (!mm || !mmget_not_zero(mm)) {
 143                 put_task_struct(priv->task);
 144                 priv->task = NULL;
 145                 return NULL;
 146         }
 147
 148         if (mmap_read_lock_killable(mm)) {
 149                 mmput(mm);
 150                 put_task_struct(priv->task);
 151                 priv->task = NULL;
 152                 return ERR_PTR(-EINTR);
 153         }
 154
 155         hold_task_mempolicy(priv);
 156         priv->tail_vma = get_gate_vma(mm);
 157
 158         vma = find_vma(mm, last_addr);
 159         if (vma)
 160                 return vma;
 161
 162         return priv->tail_vma;
 163 }
 164
 165 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
 166 {
 167         struct proc_maps_private *priv = m->private;
 168         struct vm_area_struct *next, *vma = v;
 169
 170         if (vma == priv->tail_vma)
 171                 next = NULL;
 172         else if (vma->vm_next)
 173                 next = vma->vm_next;
 174         else
 175                 next = priv->tail_vma;
 176
 177         *ppos = next ? next->vm_start : -1UL;
 178
 179         return next;
 180 }
 181
 182 static void m_stop(struct seq_file *m, void *v)
 183 {
 184         struct proc_maps_private *priv = m->private;
 185         struct mm_struct *mm = priv->mm;
 186
 187         if (!priv->task)
 188                 return;
 189
 190         release_task_mempolicy(priv);
 191         mmap_read_unlock(mm);
 192         mmput(mm);
 193         put_task_struct(priv->task);
 194         priv->task = NULL;
 195 }
 196
 197 static int proc_maps_open(struct inode *inode, struct file *file,
 198                         const struct seq_operations *ops, int psize)
 199 {
 200         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 201
 202         if (!priv)
 203                 return -ENOMEM;
 204
 205         priv->inode = inode;
 206         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 207         if (IS_ERR(priv->mm)) {
 208                 int err = PTR_ERR(priv->mm);
 209
 210                 seq_release_private(inode, file);
 211                 return err;
 212         }
 213
 214         return 0;
 215 }
 216
 217 static int proc_map_release(struct inode *inode, struct file *file)
 218 {
 219         struct seq_file *seq = file->private_data;
 220         struct proc_maps_private *priv = seq->private;
 221
 222         if (priv->mm)
 223                 mmdrop(priv->mm);
 224
 225         return seq_release_private(inode, file);
 226 }
 227
 228 static int do_maps_open(struct inode *inode, struct file *file,
 229                         const struct seq_operations *ops)
 230 {
 231         return proc_maps_open(inode, file, ops,
 232                                 sizeof(struct proc_maps_private));
 233 }
 234
 235 /*
 236  * Indicate if the VMA is a stack for the given task; for
 237  * /proc/PID/maps that is the stack of the main task.
 238  */
 239 static int is_stack(struct vm_area_struct *vma)
 240 {
 241         /*
 242          * We make no effort to guess what a given thread considers to be
 243          * its "stack".  It's not even well-defined for programs written
 244          * languages like Go.
 245          */
 246         return vma->vm_start <= vma->vm_mm->start_stack &&
 247                 vma->vm_end >= vma->vm_mm->start_stack;
 248 }
 249
 250 static void show_vma_header_prefix(struct seq_file *m,
 251                                    unsigned long start, unsigned long end,
 252                                    vm_flags_t flags, unsigned long long pgoff,
 253                                    dev_t dev, unsigned long ino)
 254 {
 255         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 256         seq_put_hex_ll(m, NULL, start, 8);
 257         seq_put_hex_ll(m, "-", end, 8);
 258         seq_putc(m, ' ');
 259         seq_putc(m, flags & VM_READ ? 'r' : '-');
 260         seq_putc(m, flags & VM_WRITE ? 'w' : '-');
 261         seq_putc(m, flags & VM_EXEC ? 'x' : '-');
 262         seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
 263         seq_put_hex_ll(m, " ", pgoff, 8);
 264         seq_put_hex_ll(m, " ", MAJOR(dev), 2);
 265         seq_put_hex_ll(m, ":", MINOR(dev), 2);
 266         seq_put_decimal_ull(m, " ", ino);
 267         seq_putc(m, ' ');
 268 }
 269
 270 static void
 271 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 272 {
 273         struct mm_struct *mm = vma->vm_mm;
 274         struct file *file = vma->vm_file;
 275         vm_flags_t flags = vma->vm_flags;
 276         unsigned long ino = 0;
 277         unsigned long long pgoff = 0;
 278         unsigned long start, end;
 279         dev_t dev = 0;
 280         const char *name = NULL;
 281
 282         if (file) {
 283                 struct inode *inode = file_inode(vma->vm_file);
 284                 dev = inode->i_sb->s_dev;
 285                 ino = inode->i_ino;
 286                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 287         }
 288
 289         start = vma->vm_start;
 290         end = vma->vm_end;
 291         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 292
 293         /*
 294          * Print the dentry name for named mappings, and a
 295          * special [heap] marker for the heap:
 296          */
 297         if (file) {
 298                 seq_pad(m, ' ');
 299                 seq_file_path(m, file, "\n");
 300                 goto done;
 301         }
 302
 303         if (vma->vm_ops && vma->vm_ops->name) {
 304                 name = vma->vm_ops->name(vma);
 305                 if (name)
 306                         goto done;
 307         }
 308
 309         name = arch_vma_name(vma);
 310         if (!name) {
 311                 if (!mm) {
 312                         name = "[vdso]";
 313                         goto done;
 314                 }
 315
 316                 if (vma->vm_start <= mm->brk &&
 317                     vma->vm_end >= mm->start_brk) {
 318                         name = "[heap]";
 319                         goto done;
 320                 }
 321
 322                 if (is_stack(vma))
 323                         name = "[stack]";
 324         }
 325
 326 done:
 327         if (name) {
 328                 seq_pad(m, ' ');
 329                 seq_puts(m, name);
 330         }
 331         seq_putc(m, '\n');
 332 }
 333
 334 static int show_map(struct seq_file *m, void *v)
 335 {
 336         show_map_vma(m, v);
 337         return 0;
 338 }
 339
 340 static const struct seq_operations proc_pid_maps_op = {
 341         .start  = m_start,
 342         .next   = m_next,
 343         .stop   = m_stop,
 344         .show   = show_map
 345 };
 346
 347 static int pid_maps_open(struct inode *inode, struct file *file)
 348 {
 349         return do_maps_open(inode, file, &proc_pid_maps_op);
 350 }
 351
 352 const struct file_operations proc_pid_maps_operations = {
 353         .open           = pid_maps_open,
 354         .read           = seq_read,
 355         .llseek         = seq_lseek,
 356         .release        = proc_map_release,
 357 };
 358
 359 /*
 360  * Proportional Set Size(PSS): my share of RSS.
 361  *
 362  * PSS of a process is the count of pages it has in memory, where each
 363  * page is divided by the number of processes sharing it.  So if a
 364  * process has 1000 pages all to itself, and 1000 shared with one other
 365  * process, its PSS will be 1500.
 366  *
 367  * To keep (accumulated) division errors low, we adopt a 64bit
 368  * fixed-point pss counter to minimize division errors. So (pss >>
 369  * PSS_SHIFT) would be the real byte count.
 370  *
 371  * A shift of 12 before division means (assuming 4K page size):
 372  *      - 1M 3-user-pages add up to 8KB errors;
 373  *      - supports mapcount up to 2^24, or 16M;
 374  *      - supports PSS up to 2^52 bytes, or 4PB.
 375  */
 376 #define PSS_SHIFT 12
 377
 378 #ifdef CONFIG_PROC_PAGE_MONITOR
 379 struct mem_size_stats {
 380         unsigned long resident;
 381         unsigned long shared_clean;
 382         unsigned long shared_dirty;
 383         unsigned long private_clean;
 384         unsigned long private_dirty;
 385         unsigned long referenced;
 386         unsigned long anonymous;
 387         unsigned long lazyfree;
 388         unsigned long anonymous_thp;
 389         unsigned long shmem_thp;
 390         unsigned long file_thp;
 391         unsigned long swap;
 392         unsigned long shared_hugetlb;
 393         unsigned long private_hugetlb;
 394         u64 pss;
 395         u64 pss_anon;
 396         u64 pss_file;
 397         u64 pss_shmem;
 398         u64 pss_locked;
 399         u64 swap_pss;
 400         bool check_shmem_swap;
 401 };
 402
 403 static void smaps_page_accumulate(struct mem_size_stats *mss,
 404                 struct page *page, unsigned long size, unsigned long pss,
 405                 bool dirty, bool locked, bool private)
 406 {
 407         mss->pss += pss;
 408
 409         if (PageAnon(page))
 410                 mss->pss_anon += pss;
 411         else if (PageSwapBacked(page))
 412                 mss->pss_shmem += pss;
 413         else
 414                 mss->pss_file += pss;
 415
 416         if (locked)
 417                 mss->pss_locked += pss;
 418
 419         if (dirty || PageDirty(page)) {
 420                 if (private)
 421                         mss->private_dirty += size;
 422                 else
 423                         mss->shared_dirty += size;
 424         } else {
 425                 if (private)
 426                         mss->private_clean += size;
 427                 else
 428                         mss->shared_clean += size;
 429         }
 430 }
 431
 432 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 433                 bool compound, bool young, bool dirty, bool locked,
 434                 bool migration)
 435 {
 436         int i, nr = compound ? compound_nr(page) : 1;
 437         unsigned long size = nr * PAGE_SIZE;
 438
 439         /*
 440          * First accumulate quantities that depend only on |size| and the type
 441          * of the compound page.
 442          */
 443         if (PageAnon(page)) {
 444                 mss->anonymous += size;
 445                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 446                         mss->lazyfree += size;
 447         }
 448
 449         mss->resident += size;
 450         /* Accumulate the size in pages that have been accessed. */
 451         if (young || page_is_young(page) || PageReferenced(page))
 452                 mss->referenced += size;
 453
 454         /*
 455          * Then accumulate quantities that may depend on sharing, or that may
 456          * differ page-by-page.
 457          *
 458          * page_count(page) == 1 guarantees the page is mapped exactly once.
 459          * If any subpage of the compound page mapped with PTE it would elevate
 460          * page_count().
 461          *
 462          * The page_mapcount() is called to get a snapshot of the mapcount.
 463          * Without holding the page lock this snapshot can be slightly wrong as
 464          * we cannot always read the mapcount atomically.  It is not safe to
 465          * call page_mapcount() even with PTL held if the page is not mapped,
 466          * especially for migration entries.  Treat regular migration entries
 467          * as mapcount == 1.
 468          */
 469         if ((page_count(page) == 1) || migration) {
 470                 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
 471                         locked, true);
 472                 return;
 473         }
 474         for (i = 0; i < nr; i++, page++) {
 475                 int mapcount = page_mapcount(page);
 476                 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
 477                 if (mapcount >= 2)
 478                         pss /= mapcount;
 479                 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
 480                                       mapcount < 2);
 481         }
 482 }
 483
 484 #ifdef CONFIG_SHMEM
 485 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 486                           __always_unused int depth, struct mm_walk *walk)
 487 {
 488         struct mem_size_stats *mss = walk->private;
 489
 490         mss->swap += shmem_partial_swap_usage(
 491                         walk->vma->vm_file->f_mapping, addr, end);
 492
 493         return 0;
 494 }
 495 #else
 496 #define smaps_pte_hole          NULL
 497 #endif /* CONFIG_SHMEM */
 498
 499 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 500                 struct mm_walk *walk)
 501 {
 502         struct mem_size_stats *mss = walk->private;
 503         struct vm_area_struct *vma = walk->vma;
 504         bool locked = !!(vma->vm_flags & VM_LOCKED);
 505         struct page *page = NULL;
 506         bool migration = false, young = false, dirty = false;
 507
 508         if (pte_present(*pte)) {
 509                 page = vm_normal_page(vma, addr, *pte);
 510                 young = pte_young(*pte);
 511                 dirty = pte_dirty(*pte);
 512         } else if (is_swap_pte(*pte)) {
 513                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 514
 515                 if (!non_swap_entry(swpent)) {
 516                         int mapcount;
 517
 518                         mss->swap += PAGE_SIZE;
 519                         mapcount = swp_swapcount(swpent);
 520                         if (mapcount >= 2) {
 521                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 522
 523                                 do_div(pss_delta, mapcount);
 524                                 mss->swap_pss += pss_delta;
 525                         } else {
 526                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 527                         }
 528                 } else if (is_migration_entry(swpent)) {
 529                         migration = true;
 530                         page = migration_entry_to_page(swpent);
 531                 } else if (is_device_private_entry(swpent))
 532                         page = device_private_entry_to_page(swpent);
 533         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 534                                                         && pte_none(*pte))) {
 535                 page = xa_load(&vma->vm_file->f_mapping->i_pages,
 536                                                 linear_page_index(vma, addr));
 537                 if (xa_is_value(page))
 538                         mss->swap += PAGE_SIZE;
 539                 return;
 540         }
 541
 542         if (!page)
 543                 return;
 544
 545         smaps_account(mss, page, false, young, dirty, locked, migration);
 546 }
 547
 548 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 549 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 550                 struct mm_walk *walk)
 551 {
 552         struct mem_size_stats *mss = walk->private;
 553         struct vm_area_struct *vma = walk->vma;
 554         bool locked = !!(vma->vm_flags & VM_LOCKED);
 555         struct page *page = NULL;
 556         bool migration = false;
 557
 558         if (pmd_present(*pmd)) {
 559                 /* FOLL_DUMP will return -EFAULT on huge zero page */
 560                 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 561         } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
 562                 swp_entry_t entry = pmd_to_swp_entry(*pmd);
 563
 564                 if (is_migration_entry(entry)) {
 565                         migration = true;
 566                         page = migration_entry_to_page(entry);
 567                 }
 568         }
 569         if (IS_ERR_OR_NULL(page))
 570                 return;
 571         if (PageAnon(page))
 572                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 573         else if (PageSwapBacked(page))
 574                 mss->shmem_thp += HPAGE_PMD_SIZE;
 575         else if (is_zone_device_page(page))
 576                 /* pass */;
 577         else
 578                 mss->file_thp += HPAGE_PMD_SIZE;
 579
 580         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
 581                       locked, migration);
 582 }
 583 #else
 584 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 585                 struct mm_walk *walk)
 586 {
 587 }
 588 #endif
 589
 590 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 591                            struct mm_walk *walk)
 592 {
 593         struct vm_area_struct *vma = walk->vma;
 594         pte_t *pte;
 595         spinlock_t *ptl;
 596
 597         ptl = pmd_trans_huge_lock(pmd, vma);
 598         if (ptl) {
 599                 smaps_pmd_entry(pmd, addr, walk);
 600                 spin_unlock(ptl);
 601                 goto out;
 602         }
 603
 604         if (pmd_trans_unstable(pmd))
 605                 goto out;
 606         /*
 607          * The mmap_lock held all the way back in m_start() is what
 608          * keeps khugepaged out of here and from collapsing things
 609          * in here.
 610          */
 611         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 612         for (; addr != end; pte++, addr += PAGE_SIZE)
 613                 smaps_pte_entry(pte, addr, walk);
 614         pte_unmap_unlock(pte - 1, ptl);
 615 out:
 616         cond_resched();
 617         return 0;
 618 }
 619
 620 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 621 {
 622         /*
 623          * Don't forget to update Documentation/ on changes.
 624          */
 625         static const char mnemonics[BITS_PER_LONG][2] = {
 626                 /*
 627                  * In case if we meet a flag we don't know about.
 628                  */
 629                 [0 ... (BITS_PER_LONG-1)] = "??",
 630
 631                 [ilog2(VM_READ)]        = "rd",
 632                 [ilog2(VM_WRITE)]       = "wr",
 633                 [ilog2(VM_EXEC)]        = "ex",
 634                 [ilog2(VM_SHARED)]      = "sh",
 635                 [ilog2(VM_MAYREAD)]     = "mr",
 636                 [ilog2(VM_MAYWRITE)]    = "mw",
 637                 [ilog2(VM_MAYEXEC)]     = "me",
 638                 [ilog2(VM_MAYSHARE)]    = "ms",
 639                 [ilog2(VM_GROWSDOWN)]   = "gd",
 640                 [ilog2(VM_PFNMAP)]      = "pf",
 641                 [ilog2(VM_DENYWRITE)]   = "dw",
 642                 [ilog2(VM_LOCKED)]      = "lo",
 643                 [ilog2(VM_IO)]          = "io",
 644                 [ilog2(VM_SEQ_READ)]    = "sr",
 645                 [ilog2(VM_RAND_READ)]   = "rr",
 646                 [ilog2(VM_DONTCOPY)]    = "dc",
 647                 [ilog2(VM_DONTEXPAND)]  = "de",
 648                 [ilog2(VM_ACCOUNT)]     = "ac",
 649                 [ilog2(VM_NORESERVE)]   = "nr",
 650                 [ilog2(VM_HUGETLB)]     = "ht",
 651                 [ilog2(VM_SYNC)]        = "sf",
 652                 [ilog2(VM_ARCH_1)]      = "ar",
 653                 [ilog2(VM_WIPEONFORK)]  = "wf",
 654                 [ilog2(VM_DONTDUMP)]    = "dd",
 655 #ifdef CONFIG_ARM64_BTI
 656                 [ilog2(VM_ARM64_BTI)]   = "bt",
 657 #endif
 658 #ifdef CONFIG_MEM_SOFT_DIRTY
 659                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 660 #endif
 661                 [ilog2(VM_MIXEDMAP)]    = "mm",
 662                 [ilog2(VM_HUGEPAGE)]    = "hg",
 663                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 664                 [ilog2(VM_MERGEABLE)]   = "mg",
 665                 [ilog2(VM_UFFD_MISSING)]= "um",
 666                 [ilog2(VM_UFFD_WP)]     = "uw",
 667 #ifdef CONFIG_ARM64_MTE
 668                 [ilog2(VM_MTE)]         = "mt",
 669                 [ilog2(VM_MTE_ALLOWED)] = "",
 670 #endif
 671 #ifdef CONFIG_ARCH_HAS_PKEYS
 672                 /* These come out via ProtectionKey: */
 673                 [ilog2(VM_PKEY_BIT0)]   = "",
 674                 [ilog2(VM_PKEY_BIT1)]   = "",
 675                 [ilog2(VM_PKEY_BIT2)]   = "",
 676                 [ilog2(VM_PKEY_BIT3)]   = "",
 677 #if VM_PKEY_BIT4
 678                 [ilog2(VM_PKEY_BIT4)]   = "",
 679 #endif
 680 #endif /* CONFIG_ARCH_HAS_PKEYS */
 681         };
 682         size_t i;
 683
 684         seq_puts(m, "VmFlags: ");
 685         for (i = 0; i < BITS_PER_LONG; i++) {
 686                 if (!mnemonics[i][0])
 687                         continue;
 688                 if (vma->vm_flags & (1UL << i)) {
 689                         seq_putc(m, mnemonics[i][0]);
 690                         seq_putc(m, mnemonics[i][1]);
 691                         seq_putc(m, ' ');
 692                 }
 693         }
 694         seq_putc(m, '\n');
 695 }
 696
 697 #ifdef CONFIG_HUGETLB_PAGE
 698 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 699                                  unsigned long addr, unsigned long end,
 700                                  struct mm_walk *walk)
 701 {
 702         struct mem_size_stats *mss = walk->private;
 703         struct vm_area_struct *vma = walk->vma;
 704         struct page *page = NULL;
 705
 706         if (pte_present(*pte)) {
 707                 page = vm_normal_page(vma, addr, *pte);
 708         } else if (is_swap_pte(*pte)) {
 709                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 710
 711                 if (is_migration_entry(swpent))
 712                         page = migration_entry_to_page(swpent);
 713                 else if (is_device_private_entry(swpent))
 714                         page = device_private_entry_to_page(swpent);
 715         }
 716         if (page) {
 717                 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
 718                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 719                 else
 720                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 721         }
 722         return 0;
 723 }
 724 #else
 725 #define smaps_hugetlb_range     NULL
 726 #endif /* HUGETLB_PAGE */
 727
 728 static const struct mm_walk_ops smaps_walk_ops = {
 729         .pmd_entry              = smaps_pte_range,
 730         .hugetlb_entry          = smaps_hugetlb_range,
 731 };
 732
 733 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 734         .pmd_entry              = smaps_pte_range,
 735         .hugetlb_entry          = smaps_hugetlb_range,
 736         .pte_hole               = smaps_pte_hole,
 737 };
 738
 739 /*
 740  * Gather mem stats from @vma with the indicated beginning
 741  * address @start, and keep them in @mss.
 742  *
 743  * Use vm_start of @vma as the beginning address if @start is 0.
 744  */
 745 static void smap_gather_stats(struct vm_area_struct *vma,
 746                 struct mem_size_stats *mss, unsigned long start)
 747 {
 748         const struct mm_walk_ops *ops = &smaps_walk_ops;
 749
 750         /* Invalid start */
 751         if (start >= vma->vm_end)
 752                 return;
 753
 754 #ifdef CONFIG_SHMEM
 755         /* In case of smaps_rollup, reset the value from previous vma */
 756         mss->check_shmem_swap = false;
 757         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 758                 /*
 759                  * For shared or readonly shmem mappings we know that all
 760                  * swapped out pages belong to the shmem object, and we can
 761                  * obtain the swap value much more efficiently. For private
 762                  * writable mappings, we might have COW pages that are
 763                  * not affected by the parent swapped out pages of the shmem
 764                  * object, so we have to distinguish them during the page walk.
 765                  * Unless we know that the shmem object (or the part mapped by
 766                  * our VMA) has no swapped out pages at all.
 767                  */
 768                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 769
 770                 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 771                                         !(vma->vm_flags & VM_WRITE))) {
 772                         mss->swap += shmem_swapped;
 773                 } else {
 774                         mss->check_shmem_swap = true;
 775                         ops = &smaps_shmem_walk_ops;
 776                 }
 777         }
 778 #endif
 779         /* mmap_lock is held in m_start */
 780         if (!start)
 781                 walk_page_vma(vma, ops, mss);
 782         else
 783                 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
 784 }
 785
 786 #define SEQ_PUT_DEC(str, val) \
 787                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 788
 789 /* Show the contents common for smaps and smaps_rollup */
 790 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 791         bool rollup_mode)
 792 {
 793         SEQ_PUT_DEC("Rss:            ", mss->resident);
 794         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 795         if (rollup_mode) {
 796                 /*
 797                  * These are meaningful only for smaps_rollup, otherwise two of
 798                  * them are zero, and the other one is the same as Pss.
 799                  */
 800                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 801                         mss->pss_anon >> PSS_SHIFT);
 802                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 803                         mss->pss_file >> PSS_SHIFT);
 804                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 805                         mss->pss_shmem >> PSS_SHIFT);
 806         }
 807         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 808         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 809         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 810         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 811         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 812         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 813         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 814         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 815         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 816         SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
 817         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 818         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 819                                   mss->private_hugetlb >> 10, 7);
 820         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 821         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 822                                         mss->swap_pss >> PSS_SHIFT);
 823         SEQ_PUT_DEC(" kB\nLocked:         ",
 824                                         mss->pss_locked >> PSS_SHIFT);
 825         seq_puts(m, " kB\n");
 826 }
 827
 828 static int show_smap(struct seq_file *m, void *v)
 829 {
 830         struct vm_area_struct *vma = v;
 831         struct mem_size_stats mss;
 832
 833         memset(&mss, 0, sizeof(mss));
 834
 835         smap_gather_stats(vma, &mss, 0);
 836
 837         show_map_vma(m, vma);
 838
 839         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 840         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 841         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 842         seq_puts(m, " kB\n");
 843
 844         __show_smap(m, &mss, false);
 845
 846         seq_printf(m, "THPeligible:    %d\n",
 847                    transparent_hugepage_active(vma));
 848
 849         if (arch_pkeys_enabled())
 850                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 851         show_smap_vma_flags(m, vma);
 852
 853         return 0;
 854 }
 855
 856 static int show_smaps_rollup(struct seq_file *m, void *v)
 857 {
 858         struct proc_maps_private *priv = m->private;
 859         struct mem_size_stats mss;
 860         struct mm_struct *mm;
 861         struct vm_area_struct *vma;
 862         unsigned long last_vma_end = 0;
 863         int ret = 0;
 864
 865         priv->task = get_proc_task(priv->inode);
 866         if (!priv->task)
 867                 return -ESRCH;
 868
 869         mm = priv->mm;
 870         if (!mm || !mmget_not_zero(mm)) {
 871                 ret = -ESRCH;
 872                 goto out_put_task;
 873         }
 874
 875         memset(&mss, 0, sizeof(mss));
 876
 877         ret = mmap_read_lock_killable(mm);
 878         if (ret)
 879                 goto out_put_mm;
 880
 881         hold_task_mempolicy(priv);
 882
 883         for (vma = priv->mm->mmap; vma;) {
 884                 smap_gather_stats(vma, &mss, 0);
 885                 last_vma_end = vma->vm_end;
 886
 887                 /*
 888                  * Release mmap_lock temporarily if someone wants to
 889                  * access it for write request.
 890                  */
 891                 if (mmap_lock_is_contended(mm)) {
 892                         mmap_read_unlock(mm);
 893                         ret = mmap_read_lock_killable(mm);
 894                         if (ret) {
 895                                 release_task_mempolicy(priv);
 896                                 goto out_put_mm;
 897                         }
 898
 899                         /*
 900                          * After dropping the lock, there are four cases to
 901                          * consider. See the following example for explanation.
 902                          *
 903                          *   +------+------+-----------+
 904                          *   | VMA1 | VMA2 | VMA3      |
 905                          *   +------+------+-----------+
 906                          *   |      |      |           |
 907                          *  4k     8k     16k         400k
 908                          *
 909                          * Suppose we drop the lock after reading VMA2 due to
 910                          * contention, then we get:
 911                          *
 912                          *      last_vma_end = 16k
 913                          *
 914                          * 1) VMA2 is freed, but VMA3 exists:
 915                          *
 916                          *    find_vma(mm, 16k - 1) will return VMA3.
 917                          *    In this case, just continue from VMA3.
 918                          *
 919                          * 2) VMA2 still exists:
 920                          *
 921                          *    find_vma(mm, 16k - 1) will return VMA2.
 922                          *    Iterate the loop like the original one.
 923                          *
 924                          * 3) No more VMAs can be found:
 925                          *
 926                          *    find_vma(mm, 16k - 1) will return NULL.
 927                          *    No more things to do, just break.
 928                          *
 929                          * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
 930                          *
 931                          *    find_vma(mm, 16k - 1) will return VMA' whose range
 932                          *    contains last_vma_end.
 933                          *    Iterate VMA' from last_vma_end.
 934                          */
 935                         vma = find_vma(mm, last_vma_end - 1);
 936                         /* Case 3 above */
 937                         if (!vma)
 938                                 break;
 939
 940                         /* Case 1 above */
 941                         if (vma->vm_start >= last_vma_end)
 942                                 continue;
 943
 944                         /* Case 4 above */
 945                         if (vma->vm_end > last_vma_end)
 946                                 smap_gather_stats(vma, &mss, last_vma_end);
 947                 }
 948                 /* Case 2 above */
 949                 vma = vma->vm_next;
 950         }
 951
 952         show_vma_header_prefix(m, priv->mm->mmap ? priv->mm->mmap->vm_start : 0,
 953                                last_vma_end, 0, 0, 0, 0);
 954         seq_pad(m, ' ');
 955         seq_puts(m, "[rollup]\n");
 956
 957         __show_smap(m, &mss, true);
 958
 959         release_task_mempolicy(priv);
 960         mmap_read_unlock(mm);
 961
 962 out_put_mm:
 963         mmput(mm);
 964 out_put_task:
 965         put_task_struct(priv->task);
 966         priv->task = NULL;
 967
 968         return ret;
 969 }
 970 #undef SEQ_PUT_DEC
 971
 972 static const struct seq_operations proc_pid_smaps_op = {
 973         .start  = m_start,
 974         .next   = m_next,
 975         .stop   = m_stop,
 976         .show   = show_smap
 977 };
 978
 979 static int pid_smaps_open(struct inode *inode, struct file *file)
 980 {
 981         return do_maps_open(inode, file, &proc_pid_smaps_op);
 982 }
 983
 984 static int smaps_rollup_open(struct inode *inode, struct file *file)
 985 {
 986         int ret;
 987         struct proc_maps_private *priv;
 988
 989         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
 990         if (!priv)
 991                 return -ENOMEM;
 992
 993         ret = single_open(file, show_smaps_rollup, priv);
 994         if (ret)
 995                 goto out_free;
 996
 997         priv->inode = inode;
 998         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 999         if (IS_ERR(priv->mm)) {
1000                 ret = PTR_ERR(priv->mm);
1001
1002                 single_release(inode, file);
1003                 goto out_free;
1004         }
1005
1006         return 0;
1007
1008 out_free:
1009         kfree(priv);
1010         return ret;
1011 }
1012
1013 static int smaps_rollup_release(struct inode *inode, struct file *file)
1014 {
1015         struct seq_file *seq = file->private_data;
1016         struct proc_maps_private *priv = seq->private;
1017
1018         if (priv->mm)
1019                 mmdrop(priv->mm);
1020
1021         kfree(priv);
1022         return single_release(inode, file);
1023 }
1024
1025 const struct file_operations proc_pid_smaps_operations = {
1026         .open           = pid_smaps_open,
1027         .read           = seq_read,
1028         .llseek         = seq_lseek,
1029         .release        = proc_map_release,
1030 };
1031
1032 const struct file_operations proc_pid_smaps_rollup_operations = {
1033         .open           = smaps_rollup_open,
1034         .read           = seq_read,
1035         .llseek         = seq_lseek,
1036         .release        = smaps_rollup_release,
1037 };
1038
1039 enum clear_refs_types {
1040         CLEAR_REFS_ALL = 1,
1041         CLEAR_REFS_ANON,
1042         CLEAR_REFS_MAPPED,
1043         CLEAR_REFS_SOFT_DIRTY,
1044         CLEAR_REFS_MM_HIWATER_RSS,
1045         CLEAR_REFS_LAST,
1046 };
1047
1048 struct clear_refs_private {
1049         enum clear_refs_types type;
1050 };
1051
1052 #ifdef CONFIG_MEM_SOFT_DIRTY
1053
1054 #define is_cow_mapping(flags) (((flags) & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE)
1055
1056 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1057 {
1058         struct page *page;
1059
1060         if (!pte_write(pte))
1061                 return false;
1062         if (!is_cow_mapping(vma->vm_flags))
1063                 return false;
1064         if (likely(!atomic_read(&vma->vm_mm->has_pinned)))
1065                 return false;
1066         page = vm_normal_page(vma, addr, pte);
1067         if (!page)
1068                 return false;
1069         return page_maybe_dma_pinned(page);
1070 }
1071
1072 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1073                 unsigned long addr, pte_t *pte)
1074 {
1075         /*
1076          * The soft-dirty tracker uses #PF-s to catch writes
1077          * to pages, so write-protect the pte as well. See the
1078          * Documentation/admin-guide/mm/soft-dirty.rst for full description
1079          * of how soft-dirty works.
1080          */
1081         pte_t ptent = *pte;
1082
1083         if (pte_present(ptent)) {
1084                 pte_t old_pte;
1085
1086                 if (pte_is_pinned(vma, addr, ptent))
1087                         return;
1088                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1089                 ptent = pte_wrprotect(old_pte);
1090                 ptent = pte_clear_soft_dirty(ptent);
1091                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1092         } else if (is_swap_pte(ptent)) {
1093                 ptent = pte_swp_clear_soft_dirty(ptent);
1094                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1095         }
1096 }
1097 #else
1098 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1099                 unsigned long addr, pte_t *pte)
1100 {
1101 }
1102 #endif
1103
1104 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1105 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1106                 unsigned long addr, pmd_t *pmdp)
1107 {
1108         pmd_t old, pmd = *pmdp;
1109
1110         if (pmd_present(pmd)) {
1111                 /* See comment in change_huge_pmd() */
1112                 old = pmdp_invalidate(vma, addr, pmdp);
1113                 if (pmd_dirty(old))
1114                         pmd = pmd_mkdirty(pmd);
1115                 if (pmd_young(old))
1116                         pmd = pmd_mkyoung(pmd);
1117
1118                 pmd = pmd_wrprotect(pmd);
1119                 pmd = pmd_clear_soft_dirty(pmd);
1120
1121                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1122         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1123                 pmd = pmd_swp_clear_soft_dirty(pmd);
1124                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1125         }
1126 }
1127 #else
1128 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1129                 unsigned long addr, pmd_t *pmdp)
1130 {
1131 }
1132 #endif
1133
1134 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1135                                 unsigned long end, struct mm_walk *walk)
1136 {
1137         struct clear_refs_private *cp = walk->private;
1138         struct vm_area_struct *vma = walk->vma;
1139         pte_t *pte, ptent;
1140         spinlock_t *ptl;
1141         struct page *page;
1142
1143         ptl = pmd_trans_huge_lock(pmd, vma);
1144         if (ptl) {
1145                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1146                         clear_soft_dirty_pmd(vma, addr, pmd);
1147                         goto out;
1148                 }
1149
1150                 if (!pmd_present(*pmd))
1151                         goto out;
1152
1153                 page = pmd_page(*pmd);
1154
1155                 /* Clear accessed and referenced bits. */
1156                 pmdp_test_and_clear_young(vma, addr, pmd);
1157                 test_and_clear_page_young(page);
1158                 ClearPageReferenced(page);
1159 out:
1160                 spin_unlock(ptl);
1161                 return 0;
1162         }
1163
1164         if (pmd_trans_unstable(pmd))
1165                 return 0;
1166
1167         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1168         for (; addr != end; pte++, addr += PAGE_SIZE) {
1169                 ptent = *pte;
1170
1171                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1172                         clear_soft_dirty(vma, addr, pte);
1173                         continue;
1174                 }
1175
1176                 if (!pte_present(ptent))
1177                         continue;
1178
1179                 page = vm_normal_page(vma, addr, ptent);
1180                 if (!page)
1181                         continue;
1182
1183                 /* Clear accessed and referenced bits. */
1184                 ptep_test_and_clear_young(vma, addr, pte);
1185                 test_and_clear_page_young(page);
1186                 ClearPageReferenced(page);
1187         }
1188         pte_unmap_unlock(pte - 1, ptl);
1189         cond_resched();
1190         return 0;
1191 }
1192
1193 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1194                                 struct mm_walk *walk)
1195 {
1196         struct clear_refs_private *cp = walk->private;
1197         struct vm_area_struct *vma = walk->vma;
1198
1199         if (vma->vm_flags & VM_PFNMAP)
1200                 return 1;
1201
1202         /*
1203          * Writing 1 to /proc/pid/clear_refs affects all pages.
1204          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1205          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1206          * Writing 4 to /proc/pid/clear_refs affects all pages.
1207          */
1208         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1209                 return 1;
1210         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1211                 return 1;
1212         return 0;
1213 }
1214
1215 static const struct mm_walk_ops clear_refs_walk_ops = {
1216         .pmd_entry              = clear_refs_pte_range,
1217         .test_walk              = clear_refs_test_walk,
1218 };
1219
1220 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1221                                 size_t count, loff_t *ppos)
1222 {
1223         struct task_struct *task;
1224         char buffer[PROC_NUMBUF];
1225         struct mm_struct *mm;
1226         struct vm_area_struct *vma;
1227         enum clear_refs_types type;
1228         int itype;
1229         int rv;
1230
1231         memset(buffer, 0, sizeof(buffer));
1232         if (count > sizeof(buffer) - 1)
1233                 count = sizeof(buffer) - 1;
1234         if (copy_from_user(buffer, buf, count))
1235                 return -EFAULT;
1236         rv = kstrtoint(strstrip(buffer), 10, &itype);
1237         if (rv < 0)
1238                 return rv;
1239         type = (enum clear_refs_types)itype;
1240         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1241                 return -EINVAL;
1242
1243         task = get_proc_task(file_inode(file));
1244         if (!task)
1245                 return -ESRCH;
1246         mm = get_task_mm(task);
1247         if (mm) {
1248                 struct mmu_notifier_range range;
1249                 struct clear_refs_private cp = {
1250                         .type = type,
1251                 };
1252
1253                 if (mmap_write_lock_killable(mm)) {
1254                         count = -EINTR;
1255                         goto out_mm;
1256                 }
1257                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1258                         /*
1259                          * Writing 5 to /proc/pid/clear_refs resets the peak
1260                          * resident set size to this mm's current rss value.
1261                          */
1262                         reset_mm_hiwater_rss(mm);
1263                         goto out_unlock;
1264                 }
1265
1266                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1267                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1268                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1269                                         continue;
1270                                 vma->vm_flags &= ~VM_SOFTDIRTY;
1271                                 vma_set_page_prot(vma);
1272                         }
1273
1274                         inc_tlb_flush_pending(mm);
1275                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1276                                                 0, NULL, mm, 0, -1UL);
1277                         mmu_notifier_invalidate_range_start(&range);
1278                 }
1279                 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops,
1280                                 &cp);
1281                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1282                         mmu_notifier_invalidate_range_end(&range);
1283                         flush_tlb_mm(mm);
1284                         dec_tlb_flush_pending(mm);
1285                 }
1286 out_unlock:
1287                 mmap_write_unlock(mm);
1288 out_mm:
1289                 mmput(mm);
1290         }
1291         put_task_struct(task);
1292
1293         return count;
1294 }
1295
1296 const struct file_operations proc_clear_refs_operations = {
1297         .write          = clear_refs_write,
1298         .llseek         = noop_llseek,
1299 };
1300
1301 typedef struct {
1302         u64 pme;
1303 } pagemap_entry_t;
1304
1305 struct pagemapread {
1306         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1307         pagemap_entry_t *buffer;
1308         bool show_pfn;
1309 };
1310
1311 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1312 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1313
1314 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1315 #define PM_PFRAME_BITS          55
1316 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1317 #define PM_SOFT_DIRTY           BIT_ULL(55)
1318 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1319 #define PM_FILE                 BIT_ULL(61)
1320 #define PM_SWAP                 BIT_ULL(62)
1321 #define PM_PRESENT              BIT_ULL(63)
1322
1323 #define PM_END_OF_BUFFER    1
1324
1325 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1326 {
1327         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1328 }
1329
1330 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1331                           struct pagemapread *pm)
1332 {
1333         pm->buffer[pm->pos++] = *pme;
1334         if (pm->pos >= pm->len)
1335                 return PM_END_OF_BUFFER;
1336         return 0;
1337 }
1338
1339 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1340                             __always_unused int depth, struct mm_walk *walk)
1341 {
1342         struct pagemapread *pm = walk->private;
1343         unsigned long addr = start;
1344         int err = 0;
1345
1346         while (addr < end) {
1347                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1348                 pagemap_entry_t pme = make_pme(0, 0);
1349                 /* End of address space hole, which we mark as non-present. */
1350                 unsigned long hole_end;
1351
1352                 if (vma)
1353                         hole_end = min(end, vma->vm_start);
1354                 else
1355                         hole_end = end;
1356
1357                 for (; addr < hole_end; addr += PAGE_SIZE) {
1358                         err = add_to_pagemap(addr, &pme, pm);
1359                         if (err)
1360                                 goto out;
1361                 }
1362
1363                 if (!vma)
1364                         break;
1365
1366                 /* Addresses in the VMA. */
1367                 if (vma->vm_flags & VM_SOFTDIRTY)
1368                         pme = make_pme(0, PM_SOFT_DIRTY);
1369                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1370                         err = add_to_pagemap(addr, &pme, pm);
1371                         if (err)
1372                                 goto out;
1373                 }
1374         }
1375 out:
1376         return err;
1377 }
1378
1379 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1380                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1381 {
1382         u64 frame = 0, flags = 0;
1383         struct page *page = NULL;
1384         bool migration = false;
1385
1386         if (pte_present(pte)) {
1387                 if (pm->show_pfn)
1388                         frame = pte_pfn(pte);
1389                 flags |= PM_PRESENT;
1390                 page = vm_normal_page(vma, addr, pte);
1391                 if (pte_soft_dirty(pte))
1392                         flags |= PM_SOFT_DIRTY;
1393         } else if (is_swap_pte(pte)) {
1394                 swp_entry_t entry;
1395                 if (pte_swp_soft_dirty(pte))
1396                         flags |= PM_SOFT_DIRTY;
1397                 entry = pte_to_swp_entry(pte);
1398                 if (pm->show_pfn)
1399                         frame = swp_type(entry) |
1400                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1401                 flags |= PM_SWAP;
1402                 if (is_migration_entry(entry)) {
1403                         migration = true;
1404                         page = migration_entry_to_page(entry);
1405                 }
1406
1407                 if (is_device_private_entry(entry))
1408                         page = device_private_entry_to_page(entry);
1409         }
1410
1411         if (page && !PageAnon(page))
1412                 flags |= PM_FILE;
1413         if (page && !migration && page_mapcount(page) == 1)
1414                 flags |= PM_MMAP_EXCLUSIVE;
1415         if (vma->vm_flags & VM_SOFTDIRTY)
1416                 flags |= PM_SOFT_DIRTY;
1417
1418         return make_pme(frame, flags);
1419 }
1420
1421 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1422                              struct mm_walk *walk)
1423 {
1424         struct vm_area_struct *vma = walk->vma;
1425         struct pagemapread *pm = walk->private;
1426         spinlock_t *ptl;
1427         pte_t *pte, *orig_pte;
1428         int err = 0;
1429 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1430         bool migration = false;
1431
1432         ptl = pmd_trans_huge_lock(pmdp, vma);
1433         if (ptl) {
1434                 u64 flags = 0, frame = 0;
1435                 pmd_t pmd = *pmdp;
1436                 struct page *page = NULL;
1437
1438                 if (vma->vm_flags & VM_SOFTDIRTY)
1439                         flags |= PM_SOFT_DIRTY;
1440
1441                 if (pmd_present(pmd)) {
1442                         page = pmd_page(pmd);
1443
1444                         flags |= PM_PRESENT;
1445                         if (pmd_soft_dirty(pmd))
1446                                 flags |= PM_SOFT_DIRTY;
1447                         if (pm->show_pfn)
1448                                 frame = pmd_pfn(pmd) +
1449                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1450                 }
1451 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1452                 else if (is_swap_pmd(pmd)) {
1453                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1454                         unsigned long offset;
1455
1456                         if (pm->show_pfn) {
1457                                 offset = swp_offset(entry) +
1458                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1459                                 frame = swp_type(entry) |
1460                                         (offset << MAX_SWAPFILES_SHIFT);
1461                         }
1462                         flags |= PM_SWAP;
1463                         if (pmd_swp_soft_dirty(pmd))
1464                                 flags |= PM_SOFT_DIRTY;
1465                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1466                         migration = is_migration_entry(entry);
1467                         page = migration_entry_to_page(entry);
1468                 }
1469 #endif
1470
1471                 if (page && !migration && page_mapcount(page) == 1)
1472                         flags |= PM_MMAP_EXCLUSIVE;
1473
1474                 for (; addr != end; addr += PAGE_SIZE) {
1475                         pagemap_entry_t pme = make_pme(frame, flags);
1476
1477                         err = add_to_pagemap(addr, &pme, pm);
1478                         if (err)
1479                                 break;
1480                         if (pm->show_pfn) {
1481                                 if (flags & PM_PRESENT)
1482                                         frame++;
1483                                 else if (flags & PM_SWAP)
1484                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1485                         }
1486                 }
1487                 spin_unlock(ptl);
1488                 return err;
1489         }
1490
1491         if (pmd_trans_unstable(pmdp))
1492                 return 0;
1493 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1494
1495         /*
1496          * We can assume that @vma always points to a valid one and @end never
1497          * goes beyond vma->vm_end.
1498          */
1499         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1500         for (; addr < end; pte++, addr += PAGE_SIZE) {
1501                 pagemap_entry_t pme;
1502
1503                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1504                 err = add_to_pagemap(addr, &pme, pm);
1505                 if (err)
1506                         break;
1507         }
1508         pte_unmap_unlock(orig_pte, ptl);
1509
1510         cond_resched();
1511
1512         return err;
1513 }
1514
1515 #ifdef CONFIG_HUGETLB_PAGE
1516 /* This function walks within one hugetlb entry in the single call */
1517 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1518                                  unsigned long addr, unsigned long end,
1519                                  struct mm_walk *walk)
1520 {
1521         struct pagemapread *pm = walk->private;
1522         struct vm_area_struct *vma = walk->vma;
1523         u64 flags = 0, frame = 0;
1524         int err = 0;
1525         pte_t pte;
1526
1527         if (vma->vm_flags & VM_SOFTDIRTY)
1528                 flags |= PM_SOFT_DIRTY;
1529
1530         pte = huge_ptep_get(ptep);
1531         if (pte_present(pte)) {
1532                 struct page *page = pte_page(pte);
1533
1534                 if (!PageAnon(page))
1535                         flags |= PM_FILE;
1536
1537                 if (page_mapcount(page) == 1)
1538                         flags |= PM_MMAP_EXCLUSIVE;
1539
1540                 flags |= PM_PRESENT;
1541                 if (pm->show_pfn)
1542                         frame = pte_pfn(pte) +
1543                                 ((addr & ~hmask) >> PAGE_SHIFT);
1544         }
1545
1546         for (; addr != end; addr += PAGE_SIZE) {
1547                 pagemap_entry_t pme = make_pme(frame, flags);
1548
1549                 err = add_to_pagemap(addr, &pme, pm);
1550                 if (err)
1551                         return err;
1552                 if (pm->show_pfn && (flags & PM_PRESENT))
1553                         frame++;
1554         }
1555
1556         cond_resched();
1557
1558         return err;
1559 }
1560 #else
1561 #define pagemap_hugetlb_range   NULL
1562 #endif /* HUGETLB_PAGE */
1563
1564 static const struct mm_walk_ops pagemap_ops = {
1565         .pmd_entry      = pagemap_pmd_range,
1566         .pte_hole       = pagemap_pte_hole,
1567         .hugetlb_entry  = pagemap_hugetlb_range,
1568 };
1569
1570 /*
1571  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1572  *
1573  * For each page in the address space, this file contains one 64-bit entry
1574  * consisting of the following:
1575  *
1576  * Bits 0-54  page frame number (PFN) if present
1577  * Bits 0-4   swap type if swapped
1578  * Bits 5-54  swap offset if swapped
1579  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1580  * Bit  56    page exclusively mapped
1581  * Bits 57-60 zero
1582  * Bit  61    page is file-page or shared-anon
1583  * Bit  62    page swapped
1584  * Bit  63    page present
1585  *
1586  * If the page is not present but in swap, then the PFN contains an
1587  * encoding of the swap file number and the page's offset into the
1588  * swap. Unmapped pages return a null PFN. This allows determining
1589  * precisely which pages are mapped (or in swap) and comparing mapped
1590  * pages between processes.
1591  *
1592  * Efficient users of this interface will use /proc/pid/maps to
1593  * determine which areas of memory are actually mapped and llseek to
1594  * skip over unmapped regions.
1595  */
1596 static ssize_t pagemap_read(struct file *file, char __user *buf,
1597                             size_t count, loff_t *ppos)
1598 {
1599         struct mm_struct *mm = file->private_data;
1600         struct pagemapread pm;
1601         unsigned long src;
1602         unsigned long svpfn;
1603         unsigned long start_vaddr;
1604         unsigned long end_vaddr;
1605         int ret = 0, copied = 0;
1606
1607         if (!mm || !mmget_not_zero(mm))
1608                 goto out;
1609
1610         ret = -EINVAL;
1611         /* file position must be aligned */
1612         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1613                 goto out_mm;
1614
1615         ret = 0;
1616         if (!count)
1617                 goto out_mm;
1618
1619         /* do not disclose physical addresses: attack vector */
1620         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1621
1622         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1623         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1624         ret = -ENOMEM;
1625         if (!pm.buffer)
1626                 goto out_mm;
1627
1628         src = *ppos;
1629         svpfn = src / PM_ENTRY_BYTES;
1630         end_vaddr = mm->task_size;
1631
1632         /* watch out for wraparound */
1633         start_vaddr = end_vaddr;
1634         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT))
1635                 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
1636
1637         /* Ensure the address is inside the task */
1638         if (start_vaddr > mm->task_size)
1639                 start_vaddr = end_vaddr;
1640
1641         /*
1642          * The odds are that this will stop walking way
1643          * before end_vaddr, because the length of the
1644          * user buffer is tracked in "pm", and the walk
1645          * will stop when we hit the end of the buffer.
1646          */
1647         ret = 0;
1648         while (count && (start_vaddr < end_vaddr)) {
1649                 int len;
1650                 unsigned long end;
1651
1652                 pm.pos = 0;
1653                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1654                 /* overflow ? */
1655                 if (end < start_vaddr || end > end_vaddr)
1656                         end = end_vaddr;
1657                 ret = mmap_read_lock_killable(mm);
1658                 if (ret)
1659                         goto out_free;
1660                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1661                 mmap_read_unlock(mm);
1662                 start_vaddr = end;
1663
1664                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1665                 if (copy_to_user(buf, pm.buffer, len)) {
1666                         ret = -EFAULT;
1667                         goto out_free;
1668                 }
1669                 copied += len;
1670                 buf += len;
1671                 count -= len;
1672         }
1673         *ppos += copied;
1674         if (!ret || ret == PM_END_OF_BUFFER)
1675                 ret = copied;
1676
1677 out_free:
1678         kfree(pm.buffer);
1679 out_mm:
1680         mmput(mm);
1681 out:
1682         return ret;
1683 }
1684
1685 static int pagemap_open(struct inode *inode, struct file *file)
1686 {
1687         struct mm_struct *mm;
1688
1689         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1690         if (IS_ERR(mm))
1691                 return PTR_ERR(mm);
1692         file->private_data = mm;
1693         return 0;
1694 }
1695
1696 static int pagemap_release(struct inode *inode, struct file *file)
1697 {
1698         struct mm_struct *mm = file->private_data;
1699
1700         if (mm)
1701                 mmdrop(mm);
1702         return 0;
1703 }
1704
1705 const struct file_operations proc_pagemap_operations = {
1706         .llseek         = mem_lseek, /* borrow this */
1707         .read           = pagemap_read,
1708         .open           = pagemap_open,
1709         .release        = pagemap_release,
1710 };
1711 #endif /* CONFIG_PROC_PAGE_MONITOR */
1712
1713 #ifdef CONFIG_NUMA
1714
1715 struct numa_maps {
1716         unsigned long pages;
1717         unsigned long anon;
1718         unsigned long active;
1719         unsigned long writeback;
1720         unsigned long mapcount_max;
1721         unsigned long dirty;
1722         unsigned long swapcache;
1723         unsigned long node[MAX_NUMNODES];
1724 };
1725
1726 struct numa_maps_private {
1727         struct proc_maps_private proc_maps;
1728         struct numa_maps md;
1729 };
1730
1731 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1732                         unsigned long nr_pages)
1733 {
1734         int count = page_mapcount(page);
1735
1736         md->pages += nr_pages;
1737         if (pte_dirty || PageDirty(page))
1738                 md->dirty += nr_pages;
1739
1740         if (PageSwapCache(page))
1741                 md->swapcache += nr_pages;
1742
1743         if (PageActive(page) || PageUnevictable(page))
1744                 md->active += nr_pages;
1745
1746         if (PageWriteback(page))
1747                 md->writeback += nr_pages;
1748
1749         if (PageAnon(page))
1750                 md->anon += nr_pages;
1751
1752         if (count > md->mapcount_max)
1753                 md->mapcount_max = count;
1754
1755         md->node[page_to_nid(page)] += nr_pages;
1756 }
1757
1758 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1759                 unsigned long addr)
1760 {
1761         struct page *page;
1762         int nid;
1763
1764         if (!pte_present(pte))
1765                 return NULL;
1766
1767         page = vm_normal_page(vma, addr, pte);
1768         if (!page)
1769                 return NULL;
1770
1771         if (PageReserved(page))
1772                 return NULL;
1773
1774         nid = page_to_nid(page);
1775         if (!node_isset(nid, node_states[N_MEMORY]))
1776                 return NULL;
1777
1778         return page;
1779 }
1780
1781 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1782 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1783                                               struct vm_area_struct *vma,
1784                                               unsigned long addr)
1785 {
1786         struct page *page;
1787         int nid;
1788
1789         if (!pmd_present(pmd))
1790                 return NULL;
1791
1792         page = vm_normal_page_pmd(vma, addr, pmd);
1793         if (!page)
1794                 return NULL;
1795
1796         if (PageReserved(page))
1797                 return NULL;
1798
1799         nid = page_to_nid(page);
1800         if (!node_isset(nid, node_states[N_MEMORY]))
1801                 return NULL;
1802
1803         return page;
1804 }
1805 #endif
1806
1807 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1808                 unsigned long end, struct mm_walk *walk)
1809 {
1810         struct numa_maps *md = walk->private;
1811         struct vm_area_struct *vma = walk->vma;
1812         spinlock_t *ptl;
1813         pte_t *orig_pte;
1814         pte_t *pte;
1815
1816 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1817         ptl = pmd_trans_huge_lock(pmd, vma);
1818         if (ptl) {
1819                 struct page *page;
1820
1821                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1822                 if (page)
1823                         gather_stats(page, md, pmd_dirty(*pmd),
1824                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1825                 spin_unlock(ptl);
1826                 return 0;
1827         }
1828
1829         if (pmd_trans_unstable(pmd))
1830                 return 0;
1831 #endif
1832         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1833         do {
1834                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1835                 if (!page)
1836                         continue;
1837                 gather_stats(page, md, pte_dirty(*pte), 1);
1838
1839         } while (pte++, addr += PAGE_SIZE, addr != end);
1840         pte_unmap_unlock(orig_pte, ptl);
1841         cond_resched();
1842         return 0;
1843 }
1844 #ifdef CONFIG_HUGETLB_PAGE
1845 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1846                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1847 {
1848         pte_t huge_pte = huge_ptep_get(pte);
1849         struct numa_maps *md;
1850         struct page *page;
1851
1852         if (!pte_present(huge_pte))
1853                 return 0;
1854
1855         page = pte_page(huge_pte);
1856         if (!page)
1857                 return 0;
1858
1859         md = walk->private;
1860         gather_stats(page, md, pte_dirty(huge_pte), 1);
1861         return 0;
1862 }
1863
1864 #else
1865 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1866                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1867 {
1868         return 0;
1869 }
1870 #endif
1871
1872 static const struct mm_walk_ops show_numa_ops = {
1873         .hugetlb_entry = gather_hugetlb_stats,
1874         .pmd_entry = gather_pte_stats,
1875 };
1876
1877 /*
1878  * Display pages allocated per node and memory policy via /proc.
1879  */
1880 static int show_numa_map(struct seq_file *m, void *v)
1881 {
1882         struct numa_maps_private *numa_priv = m->private;
1883         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1884         struct vm_area_struct *vma = v;
1885         struct numa_maps *md = &numa_priv->md;
1886         struct file *file = vma->vm_file;
1887         struct mm_struct *mm = vma->vm_mm;
1888         struct mempolicy *pol;
1889         char buffer[64];
1890         int nid;
1891
1892         if (!mm)
1893                 return 0;
1894
1895         /* Ensure we start with an empty set of numa_maps statistics. */
1896         memset(md, 0, sizeof(*md));
1897
1898         pol = __get_vma_policy(vma, vma->vm_start);
1899         if (pol) {
1900                 mpol_to_str(buffer, sizeof(buffer), pol);
1901                 mpol_cond_put(pol);
1902         } else {
1903                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1904         }
1905
1906         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1907
1908         if (file) {
1909                 seq_puts(m, " file=");
1910                 seq_file_path(m, file, "\n\t= ");
1911         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1912                 seq_puts(m, " heap");
1913         } else if (is_stack(vma)) {
1914                 seq_puts(m, " stack");
1915         }
1916
1917         if (is_vm_hugetlb_page(vma))
1918                 seq_puts(m, " huge");
1919
1920         /* mmap_lock is held by m_start */
1921         walk_page_vma(vma, &show_numa_ops, md);
1922
1923         if (!md->pages)
1924                 goto out;
1925
1926         if (md->anon)
1927                 seq_printf(m, " anon=%lu", md->anon);
1928
1929         if (md->dirty)
1930                 seq_printf(m, " dirty=%lu", md->dirty);
1931
1932         if (md->pages != md->anon && md->pages != md->dirty)
1933                 seq_printf(m, " mapped=%lu", md->pages);
1934
1935         if (md->mapcount_max > 1)
1936                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1937
1938         if (md->swapcache)
1939                 seq_printf(m, " swapcache=%lu", md->swapcache);
1940
1941         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1942                 seq_printf(m, " active=%lu", md->active);
1943
1944         if (md->writeback)
1945                 seq_printf(m, " writeback=%lu", md->writeback);
1946
1947         for_each_node_state(nid, N_MEMORY)
1948                 if (md->node[nid])
1949                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1950
1951         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1952 out:
1953         seq_putc(m, '\n');
1954         return 0;
1955 }
1956
1957 static const struct seq_operations proc_pid_numa_maps_op = {
1958         .start  = m_start,
1959         .next   = m_next,
1960         .stop   = m_stop,
1961         .show   = show_numa_map,
1962 };
1963
1964 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1965 {
1966         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1967                                 sizeof(struct numa_maps_private));
1968 }
1969
1970 const struct file_operations proc_pid_numa_maps_operations = {
1971         .open           = pid_numa_maps_open,
1972         .read           = seq_read,
1973         .llseek         = seq_lseek,
1974         .release        = proc_map_release,
1975 };
1976
1977 #endif /* CONFIG_NUMA */