fs/proc/task_mmu.c

   1 #include <linux/mm.h>
   2 #include <linux/vmacache.h>
   3 #include <linux/hugetlb.h>
   4 #include <linux/huge_mm.h>
   5 #include <linux/mount.h>
   6 #include <linux/seq_file.h>
   7 #include <linux/highmem.h>
   8 #include <linux/ptrace.h>
   9 #include <linux/slab.h>
  10 #include <linux/pagemap.h>
  11 #include <linux/mempolicy.h>
  12 #include <linux/rmap.h>
  13 #include <linux/swap.h>
  14 #include <linux/swapops.h>
  15 #include <linux/mmu_notifier.h>
  16 #include <linux/page_idle.h>
  17 #include <linux/shmem_fs.h>
  18
  19 #include <asm/elf.h>
  20 #include <asm/uaccess.h>
  21 #include <asm/tlbflush.h>
  22 #include "internal.h"
  23
  24 void task_mem(struct seq_file *m, struct mm_struct *mm)
  25 {
  26         unsigned long text, lib, swap, ptes, pmds, anon, file, shmem;
  27         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  28
  29         anon = get_mm_counter(mm, MM_ANONPAGES);
  30         file = get_mm_counter(mm, MM_FILEPAGES);
  31         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  32
  33         /*
  34          * Note: to minimize their overhead, mm maintains hiwater_vm and
  35          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  36          * collector of these hiwater stats must therefore get total_vm
  37          * and rss too, which will usually be the higher.  Barriers? not
  38          * worth the effort, such snapshots can always be inconsistent.
  39          */
  40         hiwater_vm = total_vm = mm->total_vm;
  41         if (hiwater_vm < mm->hiwater_vm)
  42                 hiwater_vm = mm->hiwater_vm;
  43         hiwater_rss = total_rss = anon + file + shmem;
  44         if (hiwater_rss < mm->hiwater_rss)
  45                 hiwater_rss = mm->hiwater_rss;
  46
  47         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  48         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  49         swap = get_mm_counter(mm, MM_SWAPENTS);
  50         ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
  51         pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
  52         seq_printf(m,
  53                 "VmPeak:\t%8lu kB\n"
  54                 "VmSize:\t%8lu kB\n"
  55                 "VmLck:\t%8lu kB\n"
  56                 "VmPin:\t%8lu kB\n"
  57                 "VmHWM:\t%8lu kB\n"
  58                 "VmRSS:\t%8lu kB\n"
  59                 "RssAnon:\t%8lu kB\n"
  60                 "RssFile:\t%8lu kB\n"
  61                 "RssShmem:\t%8lu kB\n"
  62                 "VmData:\t%8lu kB\n"
  63                 "VmStk:\t%8lu kB\n"
  64                 "VmExe:\t%8lu kB\n"
  65                 "VmLib:\t%8lu kB\n"
  66                 "VmPTE:\t%8lu kB\n"
  67                 "VmPMD:\t%8lu kB\n"
  68                 "VmSwap:\t%8lu kB\n",
  69                 hiwater_vm << (PAGE_SHIFT-10),
  70                 total_vm << (PAGE_SHIFT-10),
  71                 mm->locked_vm << (PAGE_SHIFT-10),
  72                 mm->pinned_vm << (PAGE_SHIFT-10),
  73                 hiwater_rss << (PAGE_SHIFT-10),
  74                 total_rss << (PAGE_SHIFT-10),
  75                 anon << (PAGE_SHIFT-10),
  76                 file << (PAGE_SHIFT-10),
  77                 shmem << (PAGE_SHIFT-10),
  78                 mm->data_vm << (PAGE_SHIFT-10),
  79                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  80                 ptes >> 10,
  81                 pmds >> 10,
  82                 swap << (PAGE_SHIFT-10));
  83         hugetlb_report_usage(m, mm);
  84 }
  85
  86 unsigned long task_vsize(struct mm_struct *mm)
  87 {
  88         return PAGE_SIZE * mm->total_vm;
  89 }
  90
  91 unsigned long task_statm(struct mm_struct *mm,
  92                          unsigned long *shared, unsigned long *text,
  93                          unsigned long *data, unsigned long *resident)
  94 {
  95         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  96                         get_mm_counter(mm, MM_SHMEMPAGES);
  97         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  98                                                                 >> PAGE_SHIFT;
  99         *data = mm->data_vm + mm->stack_vm;
 100         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
 101         return mm->total_vm;
 102 }
 103
 104 #ifdef CONFIG_NUMA
 105 /*
 106  * Save get_task_policy() for show_numa_map().
 107  */
 108 static void hold_task_mempolicy(struct proc_maps_private *priv)
 109 {
 110         struct task_struct *task = priv->task;
 111
 112         task_lock(task);
 113         priv->task_mempolicy = get_task_policy(task);
 114         mpol_get(priv->task_mempolicy);
 115         task_unlock(task);
 116 }
 117 static void release_task_mempolicy(struct proc_maps_private *priv)
 118 {
 119         mpol_put(priv->task_mempolicy);
 120 }
 121 #else
 122 static void hold_task_mempolicy(struct proc_maps_private *priv)
 123 {
 124 }
 125 static void release_task_mempolicy(struct proc_maps_private *priv)
 126 {
 127 }
 128 #endif
 129
 130 static void vma_stop(struct proc_maps_private *priv)
 131 {
 132         struct mm_struct *mm = priv->mm;
 133
 134         release_task_mempolicy(priv);
 135         up_read(&mm->mmap_sem);
 136         mmput(mm);
 137 }
 138
 139 static struct vm_area_struct *
 140 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
 141 {
 142         if (vma == priv->tail_vma)
 143                 return NULL;
 144         return vma->vm_next ?: priv->tail_vma;
 145 }
 146
 147 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
 148 {
 149         if (m->count < m->size) /* vma is copied successfully */
 150                 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
 151 }
 152
 153 static void *m_start(struct seq_file *m, loff_t *ppos)
 154 {
 155         struct proc_maps_private *priv = m->private;
 156         unsigned long last_addr = m->version;
 157         struct mm_struct *mm;
 158         struct vm_area_struct *vma;
 159         unsigned int pos = *ppos;
 160
 161         /* See m_cache_vma(). Zero at the start or after lseek. */
 162         if (last_addr == -1UL)
 163                 return NULL;
 164
 165         priv->task = get_proc_task(priv->inode);
 166         if (!priv->task)
 167                 return ERR_PTR(-ESRCH);
 168
 169         mm = priv->mm;
 170         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
 171                 return NULL;
 172
 173         down_read(&mm->mmap_sem);
 174         hold_task_mempolicy(priv);
 175         priv->tail_vma = get_gate_vma(mm);
 176
 177         if (last_addr) {
 178                 vma = find_vma(mm, last_addr);
 179                 if (vma && (vma = m_next_vma(priv, vma)))
 180                         return vma;
 181         }
 182
 183         m->version = 0;
 184         if (pos < mm->map_count) {
 185                 for (vma = mm->mmap; pos; pos--) {
 186                         m->version = vma->vm_start;
 187                         vma = vma->vm_next;
 188                 }
 189                 return vma;
 190         }
 191
 192         /* we do not bother to update m->version in this case */
 193         if (pos == mm->map_count && priv->tail_vma)
 194                 return priv->tail_vma;
 195
 196         vma_stop(priv);
 197         return NULL;
 198 }
 199
 200 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 201 {
 202         struct proc_maps_private *priv = m->private;
 203         struct vm_area_struct *next;
 204
 205         (*pos)++;
 206         next = m_next_vma(priv, v);
 207         if (!next)
 208                 vma_stop(priv);
 209         return next;
 210 }
 211
 212 static void m_stop(struct seq_file *m, void *v)
 213 {
 214         struct proc_maps_private *priv = m->private;
 215
 216         if (!IS_ERR_OR_NULL(v))
 217                 vma_stop(priv);
 218         if (priv->task) {
 219                 put_task_struct(priv->task);
 220                 priv->task = NULL;
 221         }
 222 }
 223
 224 static int proc_maps_open(struct inode *inode, struct file *file,
 225                         const struct seq_operations *ops, int psize)
 226 {
 227         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 228
 229         if (!priv)
 230                 return -ENOMEM;
 231
 232         priv->inode = inode;
 233         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 234         if (IS_ERR(priv->mm)) {
 235                 int err = PTR_ERR(priv->mm);
 236
 237                 seq_release_private(inode, file);
 238                 return err;
 239         }
 240
 241         return 0;
 242 }
 243
 244 static int proc_map_release(struct inode *inode, struct file *file)
 245 {
 246         struct seq_file *seq = file->private_data;
 247         struct proc_maps_private *priv = seq->private;
 248
 249         if (priv->mm)
 250                 mmdrop(priv->mm);
 251
 252         return seq_release_private(inode, file);
 253 }
 254
 255 static int do_maps_open(struct inode *inode, struct file *file,
 256                         const struct seq_operations *ops)
 257 {
 258         return proc_maps_open(inode, file, ops,
 259                                 sizeof(struct proc_maps_private));
 260 }
 261
 262 /*
 263  * Indicate if the VMA is a stack for the given task; for
 264  * /proc/PID/maps that is the stack of the main task.
 265  */
 266 static int is_stack(struct proc_maps_private *priv,
 267                     struct vm_area_struct *vma, int is_pid)
 268 {
 269         int stack = 0;
 270
 271         if (is_pid) {
 272                 stack = vma->vm_start <= vma->vm_mm->start_stack &&
 273                         vma->vm_end >= vma->vm_mm->start_stack;
 274         } else {
 275                 struct inode *inode = priv->inode;
 276                 struct task_struct *task;
 277
 278                 rcu_read_lock();
 279                 task = pid_task(proc_pid(inode), PIDTYPE_PID);
 280                 if (task)
 281                         stack = vma_is_stack_for_task(vma, task);
 282                 rcu_read_unlock();
 283         }
 284         return stack;
 285 }
 286
 287 static void
 288 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 289 {
 290         struct mm_struct *mm = vma->vm_mm;
 291         struct file *file = vma->vm_file;
 292         struct proc_maps_private *priv = m->private;
 293         vm_flags_t flags = vma->vm_flags;
 294         unsigned long ino = 0;
 295         unsigned long long pgoff = 0;
 296         unsigned long start, end;
 297         dev_t dev = 0;
 298         const char *name = NULL;
 299
 300         if (file) {
 301                 struct inode *inode = file_inode(vma->vm_file);
 302                 dev = inode->i_sb->s_dev;
 303                 ino = inode->i_ino;
 304                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 305         }
 306
 307         /* We don't show the stack guard page in /proc/maps */
 308         start = vma->vm_start;
 309         if (stack_guard_page_start(vma, start))
 310                 start += PAGE_SIZE;
 311         end = vma->vm_end;
 312         if (stack_guard_page_end(vma, end))
 313                 end -= PAGE_SIZE;
 314
 315         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 316         seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
 317                         start,
 318                         end,
 319                         flags & VM_READ ? 'r' : '-',
 320                         flags & VM_WRITE ? 'w' : '-',
 321                         flags & VM_EXEC ? 'x' : '-',
 322                         flags & VM_MAYSHARE ? 's' : 'p',
 323                         pgoff,
 324                         MAJOR(dev), MINOR(dev), ino);
 325
 326         /*
 327          * Print the dentry name for named mappings, and a
 328          * special [heap] marker for the heap:
 329          */
 330         if (file) {
 331                 seq_pad(m, ' ');
 332                 seq_file_path(m, file, "\n");
 333                 goto done;
 334         }
 335
 336         if (vma->vm_ops && vma->vm_ops->name) {
 337                 name = vma->vm_ops->name(vma);
 338                 if (name)
 339                         goto done;
 340         }
 341
 342         name = arch_vma_name(vma);
 343         if (!name) {
 344                 if (!mm) {
 345                         name = "[vdso]";
 346                         goto done;
 347                 }
 348
 349                 if (vma->vm_start <= mm->brk &&
 350                     vma->vm_end >= mm->start_brk) {
 351                         name = "[heap]";
 352                         goto done;
 353                 }
 354
 355                 if (is_stack(priv, vma, is_pid))
 356                         name = "[stack]";
 357         }
 358
 359 done:
 360         if (name) {
 361                 seq_pad(m, ' ');
 362                 seq_puts(m, name);
 363         }
 364         seq_putc(m, '\n');
 365 }
 366
 367 static int show_map(struct seq_file *m, void *v, int is_pid)
 368 {
 369         show_map_vma(m, v, is_pid);
 370         m_cache_vma(m, v);
 371         return 0;
 372 }
 373
 374 static int show_pid_map(struct seq_file *m, void *v)
 375 {
 376         return show_map(m, v, 1);
 377 }
 378
 379 static int show_tid_map(struct seq_file *m, void *v)
 380 {
 381         return show_map(m, v, 0);
 382 }
 383
 384 static const struct seq_operations proc_pid_maps_op = {
 385         .start  = m_start,
 386         .next   = m_next,
 387         .stop   = m_stop,
 388         .show   = show_pid_map
 389 };
 390
 391 static const struct seq_operations proc_tid_maps_op = {
 392         .start  = m_start,
 393         .next   = m_next,
 394         .stop   = m_stop,
 395         .show   = show_tid_map
 396 };
 397
 398 static int pid_maps_open(struct inode *inode, struct file *file)
 399 {
 400         return do_maps_open(inode, file, &proc_pid_maps_op);
 401 }
 402
 403 static int tid_maps_open(struct inode *inode, struct file *file)
 404 {
 405         return do_maps_open(inode, file, &proc_tid_maps_op);
 406 }
 407
 408 const struct file_operations proc_pid_maps_operations = {
 409         .open           = pid_maps_open,
 410         .read           = seq_read,
 411         .llseek         = seq_lseek,
 412         .release        = proc_map_release,
 413 };
 414
 415 const struct file_operations proc_tid_maps_operations = {
 416         .open           = tid_maps_open,
 417         .read           = seq_read,
 418         .llseek         = seq_lseek,
 419         .release        = proc_map_release,
 420 };
 421
 422 /*
 423  * Proportional Set Size(PSS): my share of RSS.
 424  *
 425  * PSS of a process is the count of pages it has in memory, where each
 426  * page is divided by the number of processes sharing it.  So if a
 427  * process has 1000 pages all to itself, and 1000 shared with one other
 428  * process, its PSS will be 1500.
 429  *
 430  * To keep (accumulated) division errors low, we adopt a 64bit
 431  * fixed-point pss counter to minimize division errors. So (pss >>
 432  * PSS_SHIFT) would be the real byte count.
 433  *
 434  * A shift of 12 before division means (assuming 4K page size):
 435  *      - 1M 3-user-pages add up to 8KB errors;
 436  *      - supports mapcount up to 2^24, or 16M;
 437  *      - supports PSS up to 2^52 bytes, or 4PB.
 438  */
 439 #define PSS_SHIFT 12
 440
 441 #ifdef CONFIG_PROC_PAGE_MONITOR
 442 struct mem_size_stats {
 443         unsigned long resident;
 444         unsigned long shared_clean;
 445         unsigned long shared_dirty;
 446         unsigned long private_clean;
 447         unsigned long private_dirty;
 448         unsigned long referenced;
 449         unsigned long anonymous;
 450         unsigned long anonymous_thp;
 451         unsigned long shmem_thp;
 452         unsigned long swap;
 453         unsigned long shared_hugetlb;
 454         unsigned long private_hugetlb;
 455         u64 pss;
 456         u64 swap_pss;
 457         bool check_shmem_swap;
 458 };
 459
 460 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 461                 bool compound, bool young, bool dirty)
 462 {
 463         int i, nr = compound ? 1 << compound_order(page) : 1;
 464         unsigned long size = nr * PAGE_SIZE;
 465
 466         if (PageAnon(page))
 467                 mss->anonymous += size;
 468
 469         mss->resident += size;
 470         /* Accumulate the size in pages that have been accessed. */
 471         if (young || page_is_young(page) || PageReferenced(page))
 472                 mss->referenced += size;
 473
 474         /*
 475          * page_count(page) == 1 guarantees the page is mapped exactly once.
 476          * If any subpage of the compound page mapped with PTE it would elevate
 477          * page_count().
 478          */
 479         if (page_count(page) == 1) {
 480                 if (dirty || PageDirty(page))
 481                         mss->private_dirty += size;
 482                 else
 483                         mss->private_clean += size;
 484                 mss->pss += (u64)size << PSS_SHIFT;
 485                 return;
 486         }
 487
 488         for (i = 0; i < nr; i++, page++) {
 489                 int mapcount = page_mapcount(page);
 490
 491                 if (mapcount >= 2) {
 492                         if (dirty || PageDirty(page))
 493                                 mss->shared_dirty += PAGE_SIZE;
 494                         else
 495                                 mss->shared_clean += PAGE_SIZE;
 496                         mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
 497                 } else {
 498                         if (dirty || PageDirty(page))
 499                                 mss->private_dirty += PAGE_SIZE;
 500                         else
 501                                 mss->private_clean += PAGE_SIZE;
 502                         mss->pss += PAGE_SIZE << PSS_SHIFT;
 503                 }
 504         }
 505 }
 506
 507 #ifdef CONFIG_SHMEM
 508 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 509                 struct mm_walk *walk)
 510 {
 511         struct mem_size_stats *mss = walk->private;
 512
 513         mss->swap += shmem_partial_swap_usage(
 514                         walk->vma->vm_file->f_mapping, addr, end);
 515
 516         return 0;
 517 }
 518 #endif
 519
 520 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 521                 struct mm_walk *walk)
 522 {
 523         struct mem_size_stats *mss = walk->private;
 524         struct vm_area_struct *vma = walk->vma;
 525         struct page *page = NULL;
 526
 527         if (pte_present(*pte)) {
 528                 page = vm_normal_page(vma, addr, *pte);
 529         } else if (is_swap_pte(*pte)) {
 530                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 531
 532                 if (!non_swap_entry(swpent)) {
 533                         int mapcount;
 534
 535                         mss->swap += PAGE_SIZE;
 536                         mapcount = swp_swapcount(swpent);
 537                         if (mapcount >= 2) {
 538                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 539
 540                                 do_div(pss_delta, mapcount);
 541                                 mss->swap_pss += pss_delta;
 542                         } else {
 543                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 544                         }
 545                 } else if (is_migration_entry(swpent))
 546                         page = migration_entry_to_page(swpent);
 547         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 548                                                         && pte_none(*pte))) {
 549                 page = find_get_entry(vma->vm_file->f_mapping,
 550                                                 linear_page_index(vma, addr));
 551                 if (!page)
 552                         return;
 553
 554                 if (radix_tree_exceptional_entry(page))
 555                         mss->swap += PAGE_SIZE;
 556                 else
 557                         put_page(page);
 558
 559                 return;
 560         }
 561
 562         if (!page)
 563                 return;
 564
 565         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
 566 }
 567
 568 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 569 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 570                 struct mm_walk *walk)
 571 {
 572         struct mem_size_stats *mss = walk->private;
 573         struct vm_area_struct *vma = walk->vma;
 574         struct page *page;
 575
 576         /* FOLL_DUMP will return -EFAULT on huge zero page */
 577         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 578         if (IS_ERR_OR_NULL(page))
 579                 return;
 580         if (PageAnon(page))
 581                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 582         else if (PageSwapBacked(page))
 583                 mss->shmem_thp += HPAGE_PMD_SIZE;
 584         else
 585                 VM_BUG_ON_PAGE(1, page);
 586         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
 587 }
 588 #else
 589 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 590                 struct mm_walk *walk)
 591 {
 592 }
 593 #endif
 594
 595 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 596                            struct mm_walk *walk)
 597 {
 598         struct vm_area_struct *vma = walk->vma;
 599         pte_t *pte;
 600         spinlock_t *ptl;
 601
 602         ptl = pmd_trans_huge_lock(pmd, vma);
 603         if (ptl) {
 604                 smaps_pmd_entry(pmd, addr, walk);
 605                 spin_unlock(ptl);
 606                 return 0;
 607         }
 608
 609         if (pmd_trans_unstable(pmd))
 610                 return 0;
 611         /*
 612          * The mmap_sem held all the way back in m_start() is what
 613          * keeps khugepaged out of here and from collapsing things
 614          * in here.
 615          */
 616         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 617         for (; addr != end; pte++, addr += PAGE_SIZE)
 618                 smaps_pte_entry(pte, addr, walk);
 619         pte_unmap_unlock(pte - 1, ptl);
 620         cond_resched();
 621         return 0;
 622 }
 623
 624 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 625 {
 626         /*
 627          * Don't forget to update Documentation/ on changes.
 628          */
 629         static const char mnemonics[BITS_PER_LONG][2] = {
 630                 /*
 631                  * In case if we meet a flag we don't know about.
 632                  */
 633                 [0 ... (BITS_PER_LONG-1)] = "??",
 634
 635                 [ilog2(VM_READ)]        = "rd",
 636                 [ilog2(VM_WRITE)]       = "wr",
 637                 [ilog2(VM_EXEC)]        = "ex",
 638                 [ilog2(VM_SHARED)]      = "sh",
 639                 [ilog2(VM_MAYREAD)]     = "mr",
 640                 [ilog2(VM_MAYWRITE)]    = "mw",
 641                 [ilog2(VM_MAYEXEC)]     = "me",
 642                 [ilog2(VM_MAYSHARE)]    = "ms",
 643                 [ilog2(VM_GROWSDOWN)]   = "gd",
 644                 [ilog2(VM_PFNMAP)]      = "pf",
 645                 [ilog2(VM_DENYWRITE)]   = "dw",
 646 #ifdef CONFIG_X86_INTEL_MPX
 647                 [ilog2(VM_MPX)]         = "mp",
 648 #endif
 649                 [ilog2(VM_LOCKED)]      = "lo",
 650                 [ilog2(VM_IO)]          = "io",
 651                 [ilog2(VM_SEQ_READ)]    = "sr",
 652                 [ilog2(VM_RAND_READ)]   = "rr",
 653                 [ilog2(VM_DONTCOPY)]    = "dc",
 654                 [ilog2(VM_DONTEXPAND)]  = "de",
 655                 [ilog2(VM_ACCOUNT)]     = "ac",
 656                 [ilog2(VM_NORESERVE)]   = "nr",
 657                 [ilog2(VM_HUGETLB)]     = "ht",
 658                 [ilog2(VM_ARCH_1)]      = "ar",
 659                 [ilog2(VM_DONTDUMP)]    = "dd",
 660 #ifdef CONFIG_MEM_SOFT_DIRTY
 661                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 662 #endif
 663                 [ilog2(VM_MIXEDMAP)]    = "mm",
 664                 [ilog2(VM_HUGEPAGE)]    = "hg",
 665                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 666                 [ilog2(VM_MERGEABLE)]   = "mg",
 667                 [ilog2(VM_UFFD_MISSING)]= "um",
 668                 [ilog2(VM_UFFD_WP)]     = "uw",
 669 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 670                 /* These come out via ProtectionKey: */
 671                 [ilog2(VM_PKEY_BIT0)]   = "",
 672                 [ilog2(VM_PKEY_BIT1)]   = "",
 673                 [ilog2(VM_PKEY_BIT2)]   = "",
 674                 [ilog2(VM_PKEY_BIT3)]   = "",
 675 #endif
 676         };
 677         size_t i;
 678
 679         seq_puts(m, "VmFlags: ");
 680         for (i = 0; i < BITS_PER_LONG; i++) {
 681                 if (!mnemonics[i][0])
 682                         continue;
 683                 if (vma->vm_flags & (1UL << i)) {
 684                         seq_printf(m, "%c%c ",
 685                                    mnemonics[i][0], mnemonics[i][1]);
 686                 }
 687         }
 688         seq_putc(m, '\n');
 689 }
 690
 691 #ifdef CONFIG_HUGETLB_PAGE
 692 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 693                                  unsigned long addr, unsigned long end,
 694                                  struct mm_walk *walk)
 695 {
 696         struct mem_size_stats *mss = walk->private;
 697         struct vm_area_struct *vma = walk->vma;
 698         struct page *page = NULL;
 699
 700         if (pte_present(*pte)) {
 701                 page = vm_normal_page(vma, addr, *pte);
 702         } else if (is_swap_pte(*pte)) {
 703                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 704
 705                 if (is_migration_entry(swpent))
 706                         page = migration_entry_to_page(swpent);
 707         }
 708         if (page) {
 709                 int mapcount = page_mapcount(page);
 710
 711                 if (mapcount >= 2)
 712                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 713                 else
 714                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 715         }
 716         return 0;
 717 }
 718 #endif /* HUGETLB_PAGE */
 719
 720 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
 721 {
 722 }
 723
 724 static int show_smap(struct seq_file *m, void *v, int is_pid)
 725 {
 726         struct vm_area_struct *vma = v;
 727         struct mem_size_stats mss;
 728         struct mm_walk smaps_walk = {
 729                 .pmd_entry = smaps_pte_range,
 730 #ifdef CONFIG_HUGETLB_PAGE
 731                 .hugetlb_entry = smaps_hugetlb_range,
 732 #endif
 733                 .mm = vma->vm_mm,
 734                 .private = &mss,
 735         };
 736
 737         memset(&mss, 0, sizeof mss);
 738
 739 #ifdef CONFIG_SHMEM
 740         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 741                 /*
 742                  * For shared or readonly shmem mappings we know that all
 743                  * swapped out pages belong to the shmem object, and we can
 744                  * obtain the swap value much more efficiently. For private
 745                  * writable mappings, we might have COW pages that are
 746                  * not affected by the parent swapped out pages of the shmem
 747                  * object, so we have to distinguish them during the page walk.
 748                  * Unless we know that the shmem object (or the part mapped by
 749                  * our VMA) has no swapped out pages at all.
 750                  */
 751                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 752
 753                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 754                                         !(vma->vm_flags & VM_WRITE)) {
 755                         mss.swap = shmem_swapped;
 756                 } else {
 757                         mss.check_shmem_swap = true;
 758                         smaps_walk.pte_hole = smaps_pte_hole;
 759                 }
 760         }
 761 #endif
 762
 763         /* mmap_sem is held in m_start */
 764         walk_page_vma(vma, &smaps_walk);
 765
 766         show_map_vma(m, vma, is_pid);
 767
 768         seq_printf(m,
 769                    "Size:           %8lu kB\n"
 770                    "Rss:            %8lu kB\n"
 771                    "Pss:            %8lu kB\n"
 772                    "Shared_Clean:   %8lu kB\n"
 773                    "Shared_Dirty:   %8lu kB\n"
 774                    "Private_Clean:  %8lu kB\n"
 775                    "Private_Dirty:  %8lu kB\n"
 776                    "Referenced:     %8lu kB\n"
 777                    "Anonymous:      %8lu kB\n"
 778                    "AnonHugePages:  %8lu kB\n"
 779                    "ShmemPmdMapped: %8lu kB\n"
 780                    "Shared_Hugetlb: %8lu kB\n"
 781                    "Private_Hugetlb: %7lu kB\n"
 782                    "Swap:           %8lu kB\n"
 783                    "SwapPss:        %8lu kB\n"
 784                    "KernelPageSize: %8lu kB\n"
 785                    "MMUPageSize:    %8lu kB\n"
 786                    "Locked:         %8lu kB\n",
 787                    (vma->vm_end - vma->vm_start) >> 10,
 788                    mss.resident >> 10,
 789                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 790                    mss.shared_clean  >> 10,
 791                    mss.shared_dirty  >> 10,
 792                    mss.private_clean >> 10,
 793                    mss.private_dirty >> 10,
 794                    mss.referenced >> 10,
 795                    mss.anonymous >> 10,
 796                    mss.anonymous_thp >> 10,
 797                    mss.shmem_thp >> 10,
 798                    mss.shared_hugetlb >> 10,
 799                    mss.private_hugetlb >> 10,
 800                    mss.swap >> 10,
 801                    (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
 802                    vma_kernel_pagesize(vma) >> 10,
 803                    vma_mmu_pagesize(vma) >> 10,
 804                    (vma->vm_flags & VM_LOCKED) ?
 805                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 806
 807         arch_show_smap(m, vma);
 808         show_smap_vma_flags(m, vma);
 809         m_cache_vma(m, vma);
 810         return 0;
 811 }
 812
 813 static int show_pid_smap(struct seq_file *m, void *v)
 814 {
 815         return show_smap(m, v, 1);
 816 }
 817
 818 static int show_tid_smap(struct seq_file *m, void *v)
 819 {
 820         return show_smap(m, v, 0);
 821 }
 822
 823 static const struct seq_operations proc_pid_smaps_op = {
 824         .start  = m_start,
 825         .next   = m_next,
 826         .stop   = m_stop,
 827         .show   = show_pid_smap
 828 };
 829
 830 static const struct seq_operations proc_tid_smaps_op = {
 831         .start  = m_start,
 832         .next   = m_next,
 833         .stop   = m_stop,
 834         .show   = show_tid_smap
 835 };
 836
 837 static int pid_smaps_open(struct inode *inode, struct file *file)
 838 {
 839         return do_maps_open(inode, file, &proc_pid_smaps_op);
 840 }
 841
 842 static int tid_smaps_open(struct inode *inode, struct file *file)
 843 {
 844         return do_maps_open(inode, file, &proc_tid_smaps_op);
 845 }
 846
 847 const struct file_operations proc_pid_smaps_operations = {
 848         .open           = pid_smaps_open,
 849         .read           = seq_read,
 850         .llseek         = seq_lseek,
 851         .release        = proc_map_release,
 852 };
 853
 854 const struct file_operations proc_tid_smaps_operations = {
 855         .open           = tid_smaps_open,
 856         .read           = seq_read,
 857         .llseek         = seq_lseek,
 858         .release        = proc_map_release,
 859 };
 860
 861 enum clear_refs_types {
 862         CLEAR_REFS_ALL = 1,
 863         CLEAR_REFS_ANON,
 864         CLEAR_REFS_MAPPED,
 865         CLEAR_REFS_SOFT_DIRTY,
 866         CLEAR_REFS_MM_HIWATER_RSS,
 867         CLEAR_REFS_LAST,
 868 };
 869
 870 struct clear_refs_private {
 871         enum clear_refs_types type;
 872 };
 873
 874 #ifdef CONFIG_MEM_SOFT_DIRTY
 875 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 876                 unsigned long addr, pte_t *pte)
 877 {
 878         /*
 879          * The soft-dirty tracker uses #PF-s to catch writes
 880          * to pages, so write-protect the pte as well. See the
 881          * Documentation/vm/soft-dirty.txt for full description
 882          * of how soft-dirty works.
 883          */
 884         pte_t ptent = *pte;
 885
 886         if (pte_present(ptent)) {
 887                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 888                 ptent = pte_wrprotect(ptent);
 889                 ptent = pte_clear_soft_dirty(ptent);
 890                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 891         } else if (is_swap_pte(ptent)) {
 892                 ptent = pte_swp_clear_soft_dirty(ptent);
 893                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 894         }
 895 }
 896 #else
 897 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 898                 unsigned long addr, pte_t *pte)
 899 {
 900 }
 901 #endif
 902
 903 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 904 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 905                 unsigned long addr, pmd_t *pmdp)
 906 {
 907         pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
 908
 909         pmd = pmd_wrprotect(pmd);
 910         pmd = pmd_clear_soft_dirty(pmd);
 911
 912         set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 913 }
 914 #else
 915 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 916                 unsigned long addr, pmd_t *pmdp)
 917 {
 918 }
 919 #endif
 920
 921 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 922                                 unsigned long end, struct mm_walk *walk)
 923 {
 924         struct clear_refs_private *cp = walk->private;
 925         struct vm_area_struct *vma = walk->vma;
 926         pte_t *pte, ptent;
 927         spinlock_t *ptl;
 928         struct page *page;
 929
 930         ptl = pmd_trans_huge_lock(pmd, vma);
 931         if (ptl) {
 932                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 933                         clear_soft_dirty_pmd(vma, addr, pmd);
 934                         goto out;
 935                 }
 936
 937                 page = pmd_page(*pmd);
 938
 939                 /* Clear accessed and referenced bits. */
 940                 pmdp_test_and_clear_young(vma, addr, pmd);
 941                 test_and_clear_page_young(page);
 942                 ClearPageReferenced(page);
 943 out:
 944                 spin_unlock(ptl);
 945                 return 0;
 946         }
 947
 948         if (pmd_trans_unstable(pmd))
 949                 return 0;
 950
 951         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 952         for (; addr != end; pte++, addr += PAGE_SIZE) {
 953                 ptent = *pte;
 954
 955                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 956                         clear_soft_dirty(vma, addr, pte);
 957                         continue;
 958                 }
 959
 960                 if (!pte_present(ptent))
 961                         continue;
 962
 963                 page = vm_normal_page(vma, addr, ptent);
 964                 if (!page)
 965                         continue;
 966
 967                 /* Clear accessed and referenced bits. */
 968                 ptep_test_and_clear_young(vma, addr, pte);
 969                 test_and_clear_page_young(page);
 970                 ClearPageReferenced(page);
 971         }
 972         pte_unmap_unlock(pte - 1, ptl);
 973         cond_resched();
 974         return 0;
 975 }
 976
 977 static int clear_refs_test_walk(unsigned long start, unsigned long end,
 978                                 struct mm_walk *walk)
 979 {
 980         struct clear_refs_private *cp = walk->private;
 981         struct vm_area_struct *vma = walk->vma;
 982
 983         if (vma->vm_flags & VM_PFNMAP)
 984                 return 1;
 985
 986         /*
 987          * Writing 1 to /proc/pid/clear_refs affects all pages.
 988          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
 989          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
 990          * Writing 4 to /proc/pid/clear_refs affects all pages.
 991          */
 992         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
 993                 return 1;
 994         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
 995                 return 1;
 996         return 0;
 997 }
 998
 999 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1000                                 size_t count, loff_t *ppos)
1001 {
1002         struct task_struct *task;
1003         char buffer[PROC_NUMBUF];
1004         struct mm_struct *mm;
1005         struct vm_area_struct *vma;
1006         enum clear_refs_types type;
1007         int itype;
1008         int rv;
1009
1010         memset(buffer, 0, sizeof(buffer));
1011         if (count > sizeof(buffer) - 1)
1012                 count = sizeof(buffer) - 1;
1013         if (copy_from_user(buffer, buf, count))
1014                 return -EFAULT;
1015         rv = kstrtoint(strstrip(buffer), 10, &itype);
1016         if (rv < 0)
1017                 return rv;
1018         type = (enum clear_refs_types)itype;
1019         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1020                 return -EINVAL;
1021
1022         task = get_proc_task(file_inode(file));
1023         if (!task)
1024                 return -ESRCH;
1025         mm = get_task_mm(task);
1026         if (mm) {
1027                 struct clear_refs_private cp = {
1028                         .type = type,
1029                 };
1030                 struct mm_walk clear_refs_walk = {
1031                         .pmd_entry = clear_refs_pte_range,
1032                         .test_walk = clear_refs_test_walk,
1033                         .mm = mm,
1034                         .private = &cp,
1035                 };
1036
1037                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1038                         if (down_write_killable(&mm->mmap_sem)) {
1039                                 count = -EINTR;
1040                                 goto out_mm;
1041                         }
1042
1043                         /*
1044                          * Writing 5 to /proc/pid/clear_refs resets the peak
1045                          * resident set size to this mm's current rss value.
1046                          */
1047                         reset_mm_hiwater_rss(mm);
1048                         up_write(&mm->mmap_sem);
1049                         goto out_mm;
1050                 }
1051
1052                 down_read(&mm->mmap_sem);
1053                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1054                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1055                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1056                                         continue;
1057                                 up_read(&mm->mmap_sem);
1058                                 if (down_write_killable(&mm->mmap_sem)) {
1059                                         count = -EINTR;
1060                                         goto out_mm;
1061                                 }
1062                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1063                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1064                                         vma_set_page_prot(vma);
1065                                 }
1066                                 downgrade_write(&mm->mmap_sem);
1067                                 break;
1068                         }
1069                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1070                 }
1071                 walk_page_range(0, ~0UL, &clear_refs_walk);
1072                 if (type == CLEAR_REFS_SOFT_DIRTY)
1073                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1074                 flush_tlb_mm(mm);
1075                 up_read(&mm->mmap_sem);
1076 out_mm:
1077                 mmput(mm);
1078         }
1079         put_task_struct(task);
1080
1081         return count;
1082 }
1083
1084 const struct file_operations proc_clear_refs_operations = {
1085         .write          = clear_refs_write,
1086         .llseek         = noop_llseek,
1087 };
1088
1089 typedef struct {
1090         u64 pme;
1091 } pagemap_entry_t;
1092
1093 struct pagemapread {
1094         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1095         pagemap_entry_t *buffer;
1096         bool show_pfn;
1097 };
1098
1099 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1100 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1101
1102 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1103 #define PM_PFRAME_BITS          55
1104 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1105 #define PM_SOFT_DIRTY           BIT_ULL(55)
1106 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1107 #define PM_FILE                 BIT_ULL(61)
1108 #define PM_SWAP                 BIT_ULL(62)
1109 #define PM_PRESENT              BIT_ULL(63)
1110
1111 #define PM_END_OF_BUFFER    1
1112
1113 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1114 {
1115         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1116 }
1117
1118 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1119                           struct pagemapread *pm)
1120 {
1121         pm->buffer[pm->pos++] = *pme;
1122         if (pm->pos >= pm->len)
1123                 return PM_END_OF_BUFFER;
1124         return 0;
1125 }
1126
1127 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1128                                 struct mm_walk *walk)
1129 {
1130         struct pagemapread *pm = walk->private;
1131         unsigned long addr = start;
1132         int err = 0;
1133
1134         while (addr < end) {
1135                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1136                 pagemap_entry_t pme = make_pme(0, 0);
1137                 /* End of address space hole, which we mark as non-present. */
1138                 unsigned long hole_end;
1139
1140                 if (vma)
1141                         hole_end = min(end, vma->vm_start);
1142                 else
1143                         hole_end = end;
1144
1145                 for (; addr < hole_end; addr += PAGE_SIZE) {
1146                         err = add_to_pagemap(addr, &pme, pm);
1147                         if (err)
1148                                 goto out;
1149                 }
1150
1151                 if (!vma)
1152                         break;
1153
1154                 /* Addresses in the VMA. */
1155                 if (vma->vm_flags & VM_SOFTDIRTY)
1156                         pme = make_pme(0, PM_SOFT_DIRTY);
1157                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1158                         err = add_to_pagemap(addr, &pme, pm);
1159                         if (err)
1160                                 goto out;
1161                 }
1162         }
1163 out:
1164         return err;
1165 }
1166
1167 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1168                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1169 {
1170         u64 frame = 0, flags = 0;
1171         struct page *page = NULL;
1172
1173         if (pte_present(pte)) {
1174                 if (pm->show_pfn)
1175                         frame = pte_pfn(pte);
1176                 flags |= PM_PRESENT;
1177                 page = vm_normal_page(vma, addr, pte);
1178                 if (pte_soft_dirty(pte))
1179                         flags |= PM_SOFT_DIRTY;
1180         } else if (is_swap_pte(pte)) {
1181                 swp_entry_t entry;
1182                 if (pte_swp_soft_dirty(pte))
1183                         flags |= PM_SOFT_DIRTY;
1184                 entry = pte_to_swp_entry(pte);
1185                 frame = swp_type(entry) |
1186                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1187                 flags |= PM_SWAP;
1188                 if (is_migration_entry(entry))
1189                         page = migration_entry_to_page(entry);
1190         }
1191
1192         if (page && !PageAnon(page))
1193                 flags |= PM_FILE;
1194         if (page && page_mapcount(page) == 1)
1195                 flags |= PM_MMAP_EXCLUSIVE;
1196         if (vma->vm_flags & VM_SOFTDIRTY)
1197                 flags |= PM_SOFT_DIRTY;
1198
1199         return make_pme(frame, flags);
1200 }
1201
1202 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1203                              struct mm_walk *walk)
1204 {
1205         struct vm_area_struct *vma = walk->vma;
1206         struct pagemapread *pm = walk->private;
1207         spinlock_t *ptl;
1208         pte_t *pte, *orig_pte;
1209         int err = 0;
1210
1211 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1212         ptl = pmd_trans_huge_lock(pmdp, vma);
1213         if (ptl) {
1214                 u64 flags = 0, frame = 0;
1215                 pmd_t pmd = *pmdp;
1216
1217                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1218                         flags |= PM_SOFT_DIRTY;
1219
1220                 /*
1221                  * Currently pmd for thp is always present because thp
1222                  * can not be swapped-out, migrated, or HWPOISONed
1223                  * (split in such cases instead.)
1224                  * This if-check is just to prepare for future implementation.
1225                  */
1226                 if (pmd_present(pmd)) {
1227                         struct page *page = pmd_page(pmd);
1228
1229                         if (page_mapcount(page) == 1)
1230                                 flags |= PM_MMAP_EXCLUSIVE;
1231
1232                         flags |= PM_PRESENT;
1233                         if (pm->show_pfn)
1234                                 frame = pmd_pfn(pmd) +
1235                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1236                 }
1237
1238                 for (; addr != end; addr += PAGE_SIZE) {
1239                         pagemap_entry_t pme = make_pme(frame, flags);
1240
1241                         err = add_to_pagemap(addr, &pme, pm);
1242                         if (err)
1243                                 break;
1244                         if (pm->show_pfn && (flags & PM_PRESENT))
1245                                 frame++;
1246                 }
1247                 spin_unlock(ptl);
1248                 return err;
1249         }
1250
1251         if (pmd_trans_unstable(pmdp))
1252                 return 0;
1253 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1254
1255         /*
1256          * We can assume that @vma always points to a valid one and @end never
1257          * goes beyond vma->vm_end.
1258          */
1259         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1260         for (; addr < end; pte++, addr += PAGE_SIZE) {
1261                 pagemap_entry_t pme;
1262
1263                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1264                 err = add_to_pagemap(addr, &pme, pm);
1265                 if (err)
1266                         break;
1267         }
1268         pte_unmap_unlock(orig_pte, ptl);
1269
1270         cond_resched();
1271
1272         return err;
1273 }
1274
1275 #ifdef CONFIG_HUGETLB_PAGE
1276 /* This function walks within one hugetlb entry in the single call */
1277 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1278                                  unsigned long addr, unsigned long end,
1279                                  struct mm_walk *walk)
1280 {
1281         struct pagemapread *pm = walk->private;
1282         struct vm_area_struct *vma = walk->vma;
1283         u64 flags = 0, frame = 0;
1284         int err = 0;
1285         pte_t pte;
1286
1287         if (vma->vm_flags & VM_SOFTDIRTY)
1288                 flags |= PM_SOFT_DIRTY;
1289
1290         pte = huge_ptep_get(ptep);
1291         if (pte_present(pte)) {
1292                 struct page *page = pte_page(pte);
1293
1294                 if (!PageAnon(page))
1295                         flags |= PM_FILE;
1296
1297                 if (page_mapcount(page) == 1)
1298                         flags |= PM_MMAP_EXCLUSIVE;
1299
1300                 flags |= PM_PRESENT;
1301                 if (pm->show_pfn)
1302                         frame = pte_pfn(pte) +
1303                                 ((addr & ~hmask) >> PAGE_SHIFT);
1304         }
1305
1306         for (; addr != end; addr += PAGE_SIZE) {
1307                 pagemap_entry_t pme = make_pme(frame, flags);
1308
1309                 err = add_to_pagemap(addr, &pme, pm);
1310                 if (err)
1311                         return err;
1312                 if (pm->show_pfn && (flags & PM_PRESENT))
1313                         frame++;
1314         }
1315
1316         cond_resched();
1317
1318         return err;
1319 }
1320 #endif /* HUGETLB_PAGE */
1321
1322 /*
1323  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1324  *
1325  * For each page in the address space, this file contains one 64-bit entry
1326  * consisting of the following:
1327  *
1328  * Bits 0-54  page frame number (PFN) if present
1329  * Bits 0-4   swap type if swapped
1330  * Bits 5-54  swap offset if swapped
1331  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1332  * Bit  56    page exclusively mapped
1333  * Bits 57-60 zero
1334  * Bit  61    page is file-page or shared-anon
1335  * Bit  62    page swapped
1336  * Bit  63    page present
1337  *
1338  * If the page is not present but in swap, then the PFN contains an
1339  * encoding of the swap file number and the page's offset into the
1340  * swap. Unmapped pages return a null PFN. This allows determining
1341  * precisely which pages are mapped (or in swap) and comparing mapped
1342  * pages between processes.
1343  *
1344  * Efficient users of this interface will use /proc/pid/maps to
1345  * determine which areas of memory are actually mapped and llseek to
1346  * skip over unmapped regions.
1347  */
1348 static ssize_t pagemap_read(struct file *file, char __user *buf,
1349                             size_t count, loff_t *ppos)
1350 {
1351         struct mm_struct *mm = file->private_data;
1352         struct pagemapread pm;
1353         struct mm_walk pagemap_walk = {};
1354         unsigned long src;
1355         unsigned long svpfn;
1356         unsigned long start_vaddr;
1357         unsigned long end_vaddr;
1358         int ret = 0, copied = 0;
1359
1360         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1361                 goto out;
1362
1363         ret = -EINVAL;
1364         /* file position must be aligned */
1365         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1366                 goto out_mm;
1367
1368         ret = 0;
1369         if (!count)
1370                 goto out_mm;
1371
1372         /* do not disclose physical addresses: attack vector */
1373         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1374
1375         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1376         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1377         ret = -ENOMEM;
1378         if (!pm.buffer)
1379                 goto out_mm;
1380
1381         pagemap_walk.pmd_entry = pagemap_pmd_range;
1382         pagemap_walk.pte_hole = pagemap_pte_hole;
1383 #ifdef CONFIG_HUGETLB_PAGE
1384         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1385 #endif
1386         pagemap_walk.mm = mm;
1387         pagemap_walk.private = &pm;
1388
1389         src = *ppos;
1390         svpfn = src / PM_ENTRY_BYTES;
1391         start_vaddr = svpfn << PAGE_SHIFT;
1392         end_vaddr = mm->task_size;
1393
1394         /* watch out for wraparound */
1395         if (svpfn > mm->task_size >> PAGE_SHIFT)
1396                 start_vaddr = end_vaddr;
1397
1398         /*
1399          * The odds are that this will stop walking way
1400          * before end_vaddr, because the length of the
1401          * user buffer is tracked in "pm", and the walk
1402          * will stop when we hit the end of the buffer.
1403          */
1404         ret = 0;
1405         while (count && (start_vaddr < end_vaddr)) {
1406                 int len;
1407                 unsigned long end;
1408
1409                 pm.pos = 0;
1410                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1411                 /* overflow ? */
1412                 if (end < start_vaddr || end > end_vaddr)
1413                         end = end_vaddr;
1414                 down_read(&mm->mmap_sem);
1415                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1416                 up_read(&mm->mmap_sem);
1417                 start_vaddr = end;
1418
1419                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1420                 if (copy_to_user(buf, pm.buffer, len)) {
1421                         ret = -EFAULT;
1422                         goto out_free;
1423                 }
1424                 copied += len;
1425                 buf += len;
1426                 count -= len;
1427         }
1428         *ppos += copied;
1429         if (!ret || ret == PM_END_OF_BUFFER)
1430                 ret = copied;
1431
1432 out_free:
1433         kfree(pm.buffer);
1434 out_mm:
1435         mmput(mm);
1436 out:
1437         return ret;
1438 }
1439
1440 static int pagemap_open(struct inode *inode, struct file *file)
1441 {
1442         struct mm_struct *mm;
1443
1444         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1445         if (IS_ERR(mm))
1446                 return PTR_ERR(mm);
1447         file->private_data = mm;
1448         return 0;
1449 }
1450
1451 static int pagemap_release(struct inode *inode, struct file *file)
1452 {
1453         struct mm_struct *mm = file->private_data;
1454
1455         if (mm)
1456                 mmdrop(mm);
1457         return 0;
1458 }
1459
1460 const struct file_operations proc_pagemap_operations = {
1461         .llseek         = mem_lseek, /* borrow this */
1462         .read           = pagemap_read,
1463         .open           = pagemap_open,
1464         .release        = pagemap_release,
1465 };
1466 #endif /* CONFIG_PROC_PAGE_MONITOR */
1467
1468 #ifdef CONFIG_NUMA
1469
1470 struct numa_maps {
1471         unsigned long pages;
1472         unsigned long anon;
1473         unsigned long active;
1474         unsigned long writeback;
1475         unsigned long mapcount_max;
1476         unsigned long dirty;
1477         unsigned long swapcache;
1478         unsigned long node[MAX_NUMNODES];
1479 };
1480
1481 struct numa_maps_private {
1482         struct proc_maps_private proc_maps;
1483         struct numa_maps md;
1484 };
1485
1486 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1487                         unsigned long nr_pages)
1488 {
1489         int count = page_mapcount(page);
1490
1491         md->pages += nr_pages;
1492         if (pte_dirty || PageDirty(page))
1493                 md->dirty += nr_pages;
1494
1495         if (PageSwapCache(page))
1496                 md->swapcache += nr_pages;
1497
1498         if (PageActive(page) || PageUnevictable(page))
1499                 md->active += nr_pages;
1500
1501         if (PageWriteback(page))
1502                 md->writeback += nr_pages;
1503
1504         if (PageAnon(page))
1505                 md->anon += nr_pages;
1506
1507         if (count > md->mapcount_max)
1508                 md->mapcount_max = count;
1509
1510         md->node[page_to_nid(page)] += nr_pages;
1511 }
1512
1513 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1514                 unsigned long addr)
1515 {
1516         struct page *page;
1517         int nid;
1518
1519         if (!pte_present(pte))
1520                 return NULL;
1521
1522         page = vm_normal_page(vma, addr, pte);
1523         if (!page)
1524                 return NULL;
1525
1526         if (PageReserved(page))
1527                 return NULL;
1528
1529         nid = page_to_nid(page);
1530         if (!node_isset(nid, node_states[N_MEMORY]))
1531                 return NULL;
1532
1533         return page;
1534 }
1535
1536 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1537 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1538                                               struct vm_area_struct *vma,
1539                                               unsigned long addr)
1540 {
1541         struct page *page;
1542         int nid;
1543
1544         if (!pmd_present(pmd))
1545                 return NULL;
1546
1547         page = vm_normal_page_pmd(vma, addr, pmd);
1548         if (!page)
1549                 return NULL;
1550
1551         if (PageReserved(page))
1552                 return NULL;
1553
1554         nid = page_to_nid(page);
1555         if (!node_isset(nid, node_states[N_MEMORY]))
1556                 return NULL;
1557
1558         return page;
1559 }
1560 #endif
1561
1562 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1563                 unsigned long end, struct mm_walk *walk)
1564 {
1565         struct numa_maps *md = walk->private;
1566         struct vm_area_struct *vma = walk->vma;
1567         spinlock_t *ptl;
1568         pte_t *orig_pte;
1569         pte_t *pte;
1570
1571 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1572         ptl = pmd_trans_huge_lock(pmd, vma);
1573         if (ptl) {
1574                 struct page *page;
1575
1576                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1577                 if (page)
1578                         gather_stats(page, md, pmd_dirty(*pmd),
1579                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1580                 spin_unlock(ptl);
1581                 return 0;
1582         }
1583
1584         if (pmd_trans_unstable(pmd))
1585                 return 0;
1586 #endif
1587         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1588         do {
1589                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1590                 if (!page)
1591                         continue;
1592                 gather_stats(page, md, pte_dirty(*pte), 1);
1593
1594         } while (pte++, addr += PAGE_SIZE, addr != end);
1595         pte_unmap_unlock(orig_pte, ptl);
1596         return 0;
1597 }
1598 #ifdef CONFIG_HUGETLB_PAGE
1599 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1600                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1601 {
1602         pte_t huge_pte = huge_ptep_get(pte);
1603         struct numa_maps *md;
1604         struct page *page;
1605
1606         if (!pte_present(huge_pte))
1607                 return 0;
1608
1609         page = pte_page(huge_pte);
1610         if (!page)
1611                 return 0;
1612
1613         md = walk->private;
1614         gather_stats(page, md, pte_dirty(huge_pte), 1);
1615         return 0;
1616 }
1617
1618 #else
1619 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1620                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1621 {
1622         return 0;
1623 }
1624 #endif
1625
1626 /*
1627  * Display pages allocated per node and memory policy via /proc.
1628  */
1629 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1630 {
1631         struct numa_maps_private *numa_priv = m->private;
1632         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1633         struct vm_area_struct *vma = v;
1634         struct numa_maps *md = &numa_priv->md;
1635         struct file *file = vma->vm_file;
1636         struct mm_struct *mm = vma->vm_mm;
1637         struct mm_walk walk = {
1638                 .hugetlb_entry = gather_hugetlb_stats,
1639                 .pmd_entry = gather_pte_stats,
1640                 .private = md,
1641                 .mm = mm,
1642         };
1643         struct mempolicy *pol;
1644         char buffer[64];
1645         int nid;
1646
1647         if (!mm)
1648                 return 0;
1649
1650         /* Ensure we start with an empty set of numa_maps statistics. */
1651         memset(md, 0, sizeof(*md));
1652
1653         pol = __get_vma_policy(vma, vma->vm_start);
1654         if (pol) {
1655                 mpol_to_str(buffer, sizeof(buffer), pol);
1656                 mpol_cond_put(pol);
1657         } else {
1658                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1659         }
1660
1661         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1662
1663         if (file) {
1664                 seq_puts(m, " file=");
1665                 seq_file_path(m, file, "\n\t= ");
1666         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1667                 seq_puts(m, " heap");
1668         } else if (is_stack(proc_priv, vma, is_pid)) {
1669                 seq_puts(m, " stack");
1670         }
1671
1672         if (is_vm_hugetlb_page(vma))
1673                 seq_puts(m, " huge");
1674
1675         /* mmap_sem is held by m_start */
1676         walk_page_vma(vma, &walk);
1677
1678         if (!md->pages)
1679                 goto out;
1680
1681         if (md->anon)
1682                 seq_printf(m, " anon=%lu", md->anon);
1683
1684         if (md->dirty)
1685                 seq_printf(m, " dirty=%lu", md->dirty);
1686
1687         if (md->pages != md->anon && md->pages != md->dirty)
1688                 seq_printf(m, " mapped=%lu", md->pages);
1689
1690         if (md->mapcount_max > 1)
1691                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1692
1693         if (md->swapcache)
1694                 seq_printf(m, " swapcache=%lu", md->swapcache);
1695
1696         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1697                 seq_printf(m, " active=%lu", md->active);
1698
1699         if (md->writeback)
1700                 seq_printf(m, " writeback=%lu", md->writeback);
1701
1702         for_each_node_state(nid, N_MEMORY)
1703                 if (md->node[nid])
1704                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1705
1706         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1707 out:
1708         seq_putc(m, '\n');
1709         m_cache_vma(m, vma);
1710         return 0;
1711 }
1712
1713 static int show_pid_numa_map(struct seq_file *m, void *v)
1714 {
1715         return show_numa_map(m, v, 1);
1716 }
1717
1718 static int show_tid_numa_map(struct seq_file *m, void *v)
1719 {
1720         return show_numa_map(m, v, 0);
1721 }
1722
1723 static const struct seq_operations proc_pid_numa_maps_op = {
1724         .start  = m_start,
1725         .next   = m_next,
1726         .stop   = m_stop,
1727         .show   = show_pid_numa_map,
1728 };
1729
1730 static const struct seq_operations proc_tid_numa_maps_op = {
1731         .start  = m_start,
1732         .next   = m_next,
1733         .stop   = m_stop,
1734         .show   = show_tid_numa_map,
1735 };
1736
1737 static int numa_maps_open(struct inode *inode, struct file *file,
1738                           const struct seq_operations *ops)
1739 {
1740         return proc_maps_open(inode, file, ops,
1741                                 sizeof(struct numa_maps_private));
1742 }
1743
1744 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1745 {
1746         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1747 }
1748
1749 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1750 {
1751         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1752 }
1753
1754 const struct file_operations proc_pid_numa_maps_operations = {
1755         .open           = pid_numa_maps_open,
1756         .read           = seq_read,
1757         .llseek         = seq_lseek,
1758         .release        = proc_map_release,
1759 };
1760
1761 const struct file_operations proc_tid_numa_maps_operations = {
1762         .open           = tid_numa_maps_open,
1763         .read           = seq_read,
1764         .llseek         = seq_lseek,
1765         .release        = proc_map_release,
1766 };
1767 #endif /* CONFIG_NUMA */