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 swap;
 452         unsigned long shared_hugetlb;
 453         unsigned long private_hugetlb;
 454         u64 pss;
 455         u64 swap_pss;
 456         bool check_shmem_swap;
 457 };
 458
 459 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 460                 bool compound, bool young, bool dirty)
 461 {
 462         int i, nr = compound ? 1 << compound_order(page) : 1;
 463         unsigned long size = nr * PAGE_SIZE;
 464
 465         if (PageAnon(page))
 466                 mss->anonymous += size;
 467
 468         mss->resident += size;
 469         /* Accumulate the size in pages that have been accessed. */
 470         if (young || page_is_young(page) || PageReferenced(page))
 471                 mss->referenced += size;
 472
 473         /*
 474          * page_count(page) == 1 guarantees the page is mapped exactly once.
 475          * If any subpage of the compound page mapped with PTE it would elevate
 476          * page_count().
 477          */
 478         if (page_count(page) == 1) {
 479                 if (dirty || PageDirty(page))
 480                         mss->private_dirty += size;
 481                 else
 482                         mss->private_clean += size;
 483                 mss->pss += (u64)size << PSS_SHIFT;
 484                 return;
 485         }
 486
 487         for (i = 0; i < nr; i++, page++) {
 488                 int mapcount = page_mapcount(page);
 489
 490                 if (mapcount >= 2) {
 491                         if (dirty || PageDirty(page))
 492                                 mss->shared_dirty += PAGE_SIZE;
 493                         else
 494                                 mss->shared_clean += PAGE_SIZE;
 495                         mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
 496                 } else {
 497                         if (dirty || PageDirty(page))
 498                                 mss->private_dirty += PAGE_SIZE;
 499                         else
 500                                 mss->private_clean += PAGE_SIZE;
 501                         mss->pss += PAGE_SIZE << PSS_SHIFT;
 502                 }
 503         }
 504 }
 505
 506 #ifdef CONFIG_SHMEM
 507 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 508                 struct mm_walk *walk)
 509 {
 510         struct mem_size_stats *mss = walk->private;
 511
 512         mss->swap += shmem_partial_swap_usage(
 513                         walk->vma->vm_file->f_mapping, addr, end);
 514
 515         return 0;
 516 }
 517 #endif
 518
 519 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 520                 struct mm_walk *walk)
 521 {
 522         struct mem_size_stats *mss = walk->private;
 523         struct vm_area_struct *vma = walk->vma;
 524         struct page *page = NULL;
 525
 526         if (pte_present(*pte)) {
 527                 page = vm_normal_page(vma, addr, *pte);
 528         } else if (is_swap_pte(*pte)) {
 529                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 530
 531                 if (!non_swap_entry(swpent)) {
 532                         int mapcount;
 533
 534                         mss->swap += PAGE_SIZE;
 535                         mapcount = swp_swapcount(swpent);
 536                         if (mapcount >= 2) {
 537                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 538
 539                                 do_div(pss_delta, mapcount);
 540                                 mss->swap_pss += pss_delta;
 541                         } else {
 542                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 543                         }
 544                 } else if (is_migration_entry(swpent))
 545                         page = migration_entry_to_page(swpent);
 546         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 547                                                         && pte_none(*pte))) {
 548                 page = find_get_entry(vma->vm_file->f_mapping,
 549                                                 linear_page_index(vma, addr));
 550                 if (!page)
 551                         return;
 552
 553                 if (radix_tree_exceptional_entry(page))
 554                         mss->swap += PAGE_SIZE;
 555                 else
 556                         put_page(page);
 557
 558                 return;
 559         }
 560
 561         if (!page)
 562                 return;
 563
 564         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
 565 }
 566
 567 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 568 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 569                 struct mm_walk *walk)
 570 {
 571         struct mem_size_stats *mss = walk->private;
 572         struct vm_area_struct *vma = walk->vma;
 573         struct page *page;
 574
 575         /* FOLL_DUMP will return -EFAULT on huge zero page */
 576         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 577         if (IS_ERR_OR_NULL(page))
 578                 return;
 579         mss->anonymous_thp += HPAGE_PMD_SIZE;
 580         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
 581 }
 582 #else
 583 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 584                 struct mm_walk *walk)
 585 {
 586 }
 587 #endif
 588
 589 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 590                            struct mm_walk *walk)
 591 {
 592         struct vm_area_struct *vma = walk->vma;
 593         pte_t *pte;
 594         spinlock_t *ptl;
 595
 596         ptl = pmd_trans_huge_lock(pmd, vma);
 597         if (ptl) {
 598                 smaps_pmd_entry(pmd, addr, walk);
 599                 spin_unlock(ptl);
 600                 return 0;
 601         }
 602
 603         if (pmd_trans_unstable(pmd))
 604                 return 0;
 605         /*
 606          * The mmap_sem held all the way back in m_start() is what
 607          * keeps khugepaged out of here and from collapsing things
 608          * in here.
 609          */
 610         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 611         for (; addr != end; pte++, addr += PAGE_SIZE)
 612                 smaps_pte_entry(pte, addr, walk);
 613         pte_unmap_unlock(pte - 1, ptl);
 614         cond_resched();
 615         return 0;
 616 }
 617
 618 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 619 {
 620         /*
 621          * Don't forget to update Documentation/ on changes.
 622          */
 623         static const char mnemonics[BITS_PER_LONG][2] = {
 624                 /*
 625                  * In case if we meet a flag we don't know about.
 626                  */
 627                 [0 ... (BITS_PER_LONG-1)] = "??",
 628
 629                 [ilog2(VM_READ)]        = "rd",
 630                 [ilog2(VM_WRITE)]       = "wr",
 631                 [ilog2(VM_EXEC)]        = "ex",
 632                 [ilog2(VM_SHARED)]      = "sh",
 633                 [ilog2(VM_MAYREAD)]     = "mr",
 634                 [ilog2(VM_MAYWRITE)]    = "mw",
 635                 [ilog2(VM_MAYEXEC)]     = "me",
 636                 [ilog2(VM_MAYSHARE)]    = "ms",
 637                 [ilog2(VM_GROWSDOWN)]   = "gd",
 638                 [ilog2(VM_PFNMAP)]      = "pf",
 639                 [ilog2(VM_DENYWRITE)]   = "dw",
 640 #ifdef CONFIG_X86_INTEL_MPX
 641                 [ilog2(VM_MPX)]         = "mp",
 642 #endif
 643                 [ilog2(VM_LOCKED)]      = "lo",
 644                 [ilog2(VM_IO)]          = "io",
 645                 [ilog2(VM_SEQ_READ)]    = "sr",
 646                 [ilog2(VM_RAND_READ)]   = "rr",
 647                 [ilog2(VM_DONTCOPY)]    = "dc",
 648                 [ilog2(VM_DONTEXPAND)]  = "de",
 649                 [ilog2(VM_ACCOUNT)]     = "ac",
 650                 [ilog2(VM_NORESERVE)]   = "nr",
 651                 [ilog2(VM_HUGETLB)]     = "ht",
 652                 [ilog2(VM_ARCH_1)]      = "ar",
 653                 [ilog2(VM_DONTDUMP)]    = "dd",
 654 #ifdef CONFIG_MEM_SOFT_DIRTY
 655                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 656 #endif
 657                 [ilog2(VM_MIXEDMAP)]    = "mm",
 658                 [ilog2(VM_HUGEPAGE)]    = "hg",
 659                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 660                 [ilog2(VM_MERGEABLE)]   = "mg",
 661                 [ilog2(VM_UFFD_MISSING)]= "um",
 662                 [ilog2(VM_UFFD_WP)]     = "uw",
 663 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 664                 /* These come out via ProtectionKey: */
 665                 [ilog2(VM_PKEY_BIT0)]   = "",
 666                 [ilog2(VM_PKEY_BIT1)]   = "",
 667                 [ilog2(VM_PKEY_BIT2)]   = "",
 668                 [ilog2(VM_PKEY_BIT3)]   = "",
 669 #endif
 670         };
 671         size_t i;
 672
 673         seq_puts(m, "VmFlags: ");
 674         for (i = 0; i < BITS_PER_LONG; i++) {
 675                 if (!mnemonics[i][0])
 676                         continue;
 677                 if (vma->vm_flags & (1UL << i)) {
 678                         seq_printf(m, "%c%c ",
 679                                    mnemonics[i][0], mnemonics[i][1]);
 680                 }
 681         }
 682         seq_putc(m, '\n');
 683 }
 684
 685 #ifdef CONFIG_HUGETLB_PAGE
 686 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 687                                  unsigned long addr, unsigned long end,
 688                                  struct mm_walk *walk)
 689 {
 690         struct mem_size_stats *mss = walk->private;
 691         struct vm_area_struct *vma = walk->vma;
 692         struct page *page = NULL;
 693
 694         if (pte_present(*pte)) {
 695                 page = vm_normal_page(vma, addr, *pte);
 696         } else if (is_swap_pte(*pte)) {
 697                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 698
 699                 if (is_migration_entry(swpent))
 700                         page = migration_entry_to_page(swpent);
 701         }
 702         if (page) {
 703                 int mapcount = page_mapcount(page);
 704
 705                 if (mapcount >= 2)
 706                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 707                 else
 708                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 709         }
 710         return 0;
 711 }
 712 #endif /* HUGETLB_PAGE */
 713
 714 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
 715 {
 716 }
 717
 718 static int show_smap(struct seq_file *m, void *v, int is_pid)
 719 {
 720         struct vm_area_struct *vma = v;
 721         struct mem_size_stats mss;
 722         struct mm_walk smaps_walk = {
 723                 .pmd_entry = smaps_pte_range,
 724 #ifdef CONFIG_HUGETLB_PAGE
 725                 .hugetlb_entry = smaps_hugetlb_range,
 726 #endif
 727                 .mm = vma->vm_mm,
 728                 .private = &mss,
 729         };
 730
 731         memset(&mss, 0, sizeof mss);
 732
 733 #ifdef CONFIG_SHMEM
 734         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 735                 /*
 736                  * For shared or readonly shmem mappings we know that all
 737                  * swapped out pages belong to the shmem object, and we can
 738                  * obtain the swap value much more efficiently. For private
 739                  * writable mappings, we might have COW pages that are
 740                  * not affected by the parent swapped out pages of the shmem
 741                  * object, so we have to distinguish them during the page walk.
 742                  * Unless we know that the shmem object (or the part mapped by
 743                  * our VMA) has no swapped out pages at all.
 744                  */
 745                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 746
 747                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 748                                         !(vma->vm_flags & VM_WRITE)) {
 749                         mss.swap = shmem_swapped;
 750                 } else {
 751                         mss.check_shmem_swap = true;
 752                         smaps_walk.pte_hole = smaps_pte_hole;
 753                 }
 754         }
 755 #endif
 756
 757         /* mmap_sem is held in m_start */
 758         walk_page_vma(vma, &smaps_walk);
 759
 760         show_map_vma(m, vma, is_pid);
 761
 762         seq_printf(m,
 763                    "Size:           %8lu kB\n"
 764                    "Rss:            %8lu kB\n"
 765                    "Pss:            %8lu kB\n"
 766                    "Shared_Clean:   %8lu kB\n"
 767                    "Shared_Dirty:   %8lu kB\n"
 768                    "Private_Clean:  %8lu kB\n"
 769                    "Private_Dirty:  %8lu kB\n"
 770                    "Referenced:     %8lu kB\n"
 771                    "Anonymous:      %8lu kB\n"
 772                    "AnonHugePages:  %8lu kB\n"
 773                    "Shared_Hugetlb: %8lu kB\n"
 774                    "Private_Hugetlb: %7lu kB\n"
 775                    "Swap:           %8lu kB\n"
 776                    "SwapPss:        %8lu kB\n"
 777                    "KernelPageSize: %8lu kB\n"
 778                    "MMUPageSize:    %8lu kB\n"
 779                    "Locked:         %8lu kB\n",
 780                    (vma->vm_end - vma->vm_start) >> 10,
 781                    mss.resident >> 10,
 782                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 783                    mss.shared_clean  >> 10,
 784                    mss.shared_dirty  >> 10,
 785                    mss.private_clean >> 10,
 786                    mss.private_dirty >> 10,
 787                    mss.referenced >> 10,
 788                    mss.anonymous >> 10,
 789                    mss.anonymous_thp >> 10,
 790                    mss.shared_hugetlb >> 10,
 791                    mss.private_hugetlb >> 10,
 792                    mss.swap >> 10,
 793                    (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
 794                    vma_kernel_pagesize(vma) >> 10,
 795                    vma_mmu_pagesize(vma) >> 10,
 796                    (vma->vm_flags & VM_LOCKED) ?
 797                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 798
 799         arch_show_smap(m, vma);
 800         show_smap_vma_flags(m, vma);
 801         m_cache_vma(m, vma);
 802         return 0;
 803 }
 804
 805 static int show_pid_smap(struct seq_file *m, void *v)
 806 {
 807         return show_smap(m, v, 1);
 808 }
 809
 810 static int show_tid_smap(struct seq_file *m, void *v)
 811 {
 812         return show_smap(m, v, 0);
 813 }
 814
 815 static const struct seq_operations proc_pid_smaps_op = {
 816         .start  = m_start,
 817         .next   = m_next,
 818         .stop   = m_stop,
 819         .show   = show_pid_smap
 820 };
 821
 822 static const struct seq_operations proc_tid_smaps_op = {
 823         .start  = m_start,
 824         .next   = m_next,
 825         .stop   = m_stop,
 826         .show   = show_tid_smap
 827 };
 828
 829 static int pid_smaps_open(struct inode *inode, struct file *file)
 830 {
 831         return do_maps_open(inode, file, &proc_pid_smaps_op);
 832 }
 833
 834 static int tid_smaps_open(struct inode *inode, struct file *file)
 835 {
 836         return do_maps_open(inode, file, &proc_tid_smaps_op);
 837 }
 838
 839 const struct file_operations proc_pid_smaps_operations = {
 840         .open           = pid_smaps_open,
 841         .read           = seq_read,
 842         .llseek         = seq_lseek,
 843         .release        = proc_map_release,
 844 };
 845
 846 const struct file_operations proc_tid_smaps_operations = {
 847         .open           = tid_smaps_open,
 848         .read           = seq_read,
 849         .llseek         = seq_lseek,
 850         .release        = proc_map_release,
 851 };
 852
 853 enum clear_refs_types {
 854         CLEAR_REFS_ALL = 1,
 855         CLEAR_REFS_ANON,
 856         CLEAR_REFS_MAPPED,
 857         CLEAR_REFS_SOFT_DIRTY,
 858         CLEAR_REFS_MM_HIWATER_RSS,
 859         CLEAR_REFS_LAST,
 860 };
 861
 862 struct clear_refs_private {
 863         enum clear_refs_types type;
 864 };
 865
 866 #ifdef CONFIG_MEM_SOFT_DIRTY
 867 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 868                 unsigned long addr, pte_t *pte)
 869 {
 870         /*
 871          * The soft-dirty tracker uses #PF-s to catch writes
 872          * to pages, so write-protect the pte as well. See the
 873          * Documentation/vm/soft-dirty.txt for full description
 874          * of how soft-dirty works.
 875          */
 876         pte_t ptent = *pte;
 877
 878         if (pte_present(ptent)) {
 879                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 880                 ptent = pte_wrprotect(ptent);
 881                 ptent = pte_clear_soft_dirty(ptent);
 882                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 883         } else if (is_swap_pte(ptent)) {
 884                 ptent = pte_swp_clear_soft_dirty(ptent);
 885                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 886         }
 887 }
 888 #else
 889 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 890                 unsigned long addr, pte_t *pte)
 891 {
 892 }
 893 #endif
 894
 895 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 896 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 897                 unsigned long addr, pmd_t *pmdp)
 898 {
 899         pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
 900
 901         pmd = pmd_wrprotect(pmd);
 902         pmd = pmd_clear_soft_dirty(pmd);
 903
 904         set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 905 }
 906 #else
 907 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 908                 unsigned long addr, pmd_t *pmdp)
 909 {
 910 }
 911 #endif
 912
 913 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 914                                 unsigned long end, struct mm_walk *walk)
 915 {
 916         struct clear_refs_private *cp = walk->private;
 917         struct vm_area_struct *vma = walk->vma;
 918         pte_t *pte, ptent;
 919         spinlock_t *ptl;
 920         struct page *page;
 921
 922         ptl = pmd_trans_huge_lock(pmd, vma);
 923         if (ptl) {
 924                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 925                         clear_soft_dirty_pmd(vma, addr, pmd);
 926                         goto out;
 927                 }
 928
 929                 page = pmd_page(*pmd);
 930
 931                 /* Clear accessed and referenced bits. */
 932                 pmdp_test_and_clear_young(vma, addr, pmd);
 933                 test_and_clear_page_young(page);
 934                 ClearPageReferenced(page);
 935 out:
 936                 spin_unlock(ptl);
 937                 return 0;
 938         }
 939
 940         if (pmd_trans_unstable(pmd))
 941                 return 0;
 942
 943         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 944         for (; addr != end; pte++, addr += PAGE_SIZE) {
 945                 ptent = *pte;
 946
 947                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 948                         clear_soft_dirty(vma, addr, pte);
 949                         continue;
 950                 }
 951
 952                 if (!pte_present(ptent))
 953                         continue;
 954
 955                 page = vm_normal_page(vma, addr, ptent);
 956                 if (!page)
 957                         continue;
 958
 959                 /* Clear accessed and referenced bits. */
 960                 ptep_test_and_clear_young(vma, addr, pte);
 961                 test_and_clear_page_young(page);
 962                 ClearPageReferenced(page);
 963         }
 964         pte_unmap_unlock(pte - 1, ptl);
 965         cond_resched();
 966         return 0;
 967 }
 968
 969 static int clear_refs_test_walk(unsigned long start, unsigned long end,
 970                                 struct mm_walk *walk)
 971 {
 972         struct clear_refs_private *cp = walk->private;
 973         struct vm_area_struct *vma = walk->vma;
 974
 975         if (vma->vm_flags & VM_PFNMAP)
 976                 return 1;
 977
 978         /*
 979          * Writing 1 to /proc/pid/clear_refs affects all pages.
 980          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
 981          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
 982          * Writing 4 to /proc/pid/clear_refs affects all pages.
 983          */
 984         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
 985                 return 1;
 986         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
 987                 return 1;
 988         return 0;
 989 }
 990
 991 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 992                                 size_t count, loff_t *ppos)
 993 {
 994         struct task_struct *task;
 995         char buffer[PROC_NUMBUF];
 996         struct mm_struct *mm;
 997         struct vm_area_struct *vma;
 998         enum clear_refs_types type;
 999         int itype;
1000         int rv;
1001
1002         memset(buffer, 0, sizeof(buffer));
1003         if (count > sizeof(buffer) - 1)
1004                 count = sizeof(buffer) - 1;
1005         if (copy_from_user(buffer, buf, count))
1006                 return -EFAULT;
1007         rv = kstrtoint(strstrip(buffer), 10, &itype);
1008         if (rv < 0)
1009                 return rv;
1010         type = (enum clear_refs_types)itype;
1011         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1012                 return -EINVAL;
1013
1014         task = get_proc_task(file_inode(file));
1015         if (!task)
1016                 return -ESRCH;
1017         mm = get_task_mm(task);
1018         if (mm) {
1019                 struct clear_refs_private cp = {
1020                         .type = type,
1021                 };
1022                 struct mm_walk clear_refs_walk = {
1023                         .pmd_entry = clear_refs_pte_range,
1024                         .test_walk = clear_refs_test_walk,
1025                         .mm = mm,
1026                         .private = &cp,
1027                 };
1028
1029                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1030                         /*
1031                          * Writing 5 to /proc/pid/clear_refs resets the peak
1032                          * resident set size to this mm's current rss value.
1033                          */
1034                         down_write(&mm->mmap_sem);
1035                         reset_mm_hiwater_rss(mm);
1036                         up_write(&mm->mmap_sem);
1037                         goto out_mm;
1038                 }
1039
1040                 down_read(&mm->mmap_sem);
1041                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1042                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1043                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1044                                         continue;
1045                                 up_read(&mm->mmap_sem);
1046                                 down_write(&mm->mmap_sem);
1047                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1048                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1049                                         vma_set_page_prot(vma);
1050                                 }
1051                                 downgrade_write(&mm->mmap_sem);
1052                                 break;
1053                         }
1054                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1055                 }
1056                 walk_page_range(0, ~0UL, &clear_refs_walk);
1057                 if (type == CLEAR_REFS_SOFT_DIRTY)
1058                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1059                 flush_tlb_mm(mm);
1060                 up_read(&mm->mmap_sem);
1061 out_mm:
1062                 mmput(mm);
1063         }
1064         put_task_struct(task);
1065
1066         return count;
1067 }
1068
1069 const struct file_operations proc_clear_refs_operations = {
1070         .write          = clear_refs_write,
1071         .llseek         = noop_llseek,
1072 };
1073
1074 typedef struct {
1075         u64 pme;
1076 } pagemap_entry_t;
1077
1078 struct pagemapread {
1079         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1080         pagemap_entry_t *buffer;
1081         bool show_pfn;
1082 };
1083
1084 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1085 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1086
1087 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1088 #define PM_PFRAME_BITS          55
1089 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1090 #define PM_SOFT_DIRTY           BIT_ULL(55)
1091 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1092 #define PM_FILE                 BIT_ULL(61)
1093 #define PM_SWAP                 BIT_ULL(62)
1094 #define PM_PRESENT              BIT_ULL(63)
1095
1096 #define PM_END_OF_BUFFER    1
1097
1098 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1099 {
1100         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1101 }
1102
1103 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1104                           struct pagemapread *pm)
1105 {
1106         pm->buffer[pm->pos++] = *pme;
1107         if (pm->pos >= pm->len)
1108                 return PM_END_OF_BUFFER;
1109         return 0;
1110 }
1111
1112 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1113                                 struct mm_walk *walk)
1114 {
1115         struct pagemapread *pm = walk->private;
1116         unsigned long addr = start;
1117         int err = 0;
1118
1119         while (addr < end) {
1120                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1121                 pagemap_entry_t pme = make_pme(0, 0);
1122                 /* End of address space hole, which we mark as non-present. */
1123                 unsigned long hole_end;
1124
1125                 if (vma)
1126                         hole_end = min(end, vma->vm_start);
1127                 else
1128                         hole_end = end;
1129
1130                 for (; addr < hole_end; addr += PAGE_SIZE) {
1131                         err = add_to_pagemap(addr, &pme, pm);
1132                         if (err)
1133                                 goto out;
1134                 }
1135
1136                 if (!vma)
1137                         break;
1138
1139                 /* Addresses in the VMA. */
1140                 if (vma->vm_flags & VM_SOFTDIRTY)
1141                         pme = make_pme(0, PM_SOFT_DIRTY);
1142                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1143                         err = add_to_pagemap(addr, &pme, pm);
1144                         if (err)
1145                                 goto out;
1146                 }
1147         }
1148 out:
1149         return err;
1150 }
1151
1152 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1153                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1154 {
1155         u64 frame = 0, flags = 0;
1156         struct page *page = NULL;
1157
1158         if (pte_present(pte)) {
1159                 if (pm->show_pfn)
1160                         frame = pte_pfn(pte);
1161                 flags |= PM_PRESENT;
1162                 page = vm_normal_page(vma, addr, pte);
1163                 if (pte_soft_dirty(pte))
1164                         flags |= PM_SOFT_DIRTY;
1165         } else if (is_swap_pte(pte)) {
1166                 swp_entry_t entry;
1167                 if (pte_swp_soft_dirty(pte))
1168                         flags |= PM_SOFT_DIRTY;
1169                 entry = pte_to_swp_entry(pte);
1170                 frame = swp_type(entry) |
1171                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1172                 flags |= PM_SWAP;
1173                 if (is_migration_entry(entry))
1174                         page = migration_entry_to_page(entry);
1175         }
1176
1177         if (page && !PageAnon(page))
1178                 flags |= PM_FILE;
1179         if (page && page_mapcount(page) == 1)
1180                 flags |= PM_MMAP_EXCLUSIVE;
1181         if (vma->vm_flags & VM_SOFTDIRTY)
1182                 flags |= PM_SOFT_DIRTY;
1183
1184         return make_pme(frame, flags);
1185 }
1186
1187 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1188                              struct mm_walk *walk)
1189 {
1190         struct vm_area_struct *vma = walk->vma;
1191         struct pagemapread *pm = walk->private;
1192         spinlock_t *ptl;
1193         pte_t *pte, *orig_pte;
1194         int err = 0;
1195
1196 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1197         ptl = pmd_trans_huge_lock(pmdp, vma);
1198         if (ptl) {
1199                 u64 flags = 0, frame = 0;
1200                 pmd_t pmd = *pmdp;
1201
1202                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1203                         flags |= PM_SOFT_DIRTY;
1204
1205                 /*
1206                  * Currently pmd for thp is always present because thp
1207                  * can not be swapped-out, migrated, or HWPOISONed
1208                  * (split in such cases instead.)
1209                  * This if-check is just to prepare for future implementation.
1210                  */
1211                 if (pmd_present(pmd)) {
1212                         struct page *page = pmd_page(pmd);
1213
1214                         if (page_mapcount(page) == 1)
1215                                 flags |= PM_MMAP_EXCLUSIVE;
1216
1217                         flags |= PM_PRESENT;
1218                         if (pm->show_pfn)
1219                                 frame = pmd_pfn(pmd) +
1220                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1221                 }
1222
1223                 for (; addr != end; addr += PAGE_SIZE) {
1224                         pagemap_entry_t pme = make_pme(frame, flags);
1225
1226                         err = add_to_pagemap(addr, &pme, pm);
1227                         if (err)
1228                                 break;
1229                         if (pm->show_pfn && (flags & PM_PRESENT))
1230                                 frame++;
1231                 }
1232                 spin_unlock(ptl);
1233                 return err;
1234         }
1235
1236         if (pmd_trans_unstable(pmdp))
1237                 return 0;
1238 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1239
1240         /*
1241          * We can assume that @vma always points to a valid one and @end never
1242          * goes beyond vma->vm_end.
1243          */
1244         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1245         for (; addr < end; pte++, addr += PAGE_SIZE) {
1246                 pagemap_entry_t pme;
1247
1248                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1249                 err = add_to_pagemap(addr, &pme, pm);
1250                 if (err)
1251                         break;
1252         }
1253         pte_unmap_unlock(orig_pte, ptl);
1254
1255         cond_resched();
1256
1257         return err;
1258 }
1259
1260 #ifdef CONFIG_HUGETLB_PAGE
1261 /* This function walks within one hugetlb entry in the single call */
1262 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1263                                  unsigned long addr, unsigned long end,
1264                                  struct mm_walk *walk)
1265 {
1266         struct pagemapread *pm = walk->private;
1267         struct vm_area_struct *vma = walk->vma;
1268         u64 flags = 0, frame = 0;
1269         int err = 0;
1270         pte_t pte;
1271
1272         if (vma->vm_flags & VM_SOFTDIRTY)
1273                 flags |= PM_SOFT_DIRTY;
1274
1275         pte = huge_ptep_get(ptep);
1276         if (pte_present(pte)) {
1277                 struct page *page = pte_page(pte);
1278
1279                 if (!PageAnon(page))
1280                         flags |= PM_FILE;
1281
1282                 if (page_mapcount(page) == 1)
1283                         flags |= PM_MMAP_EXCLUSIVE;
1284
1285                 flags |= PM_PRESENT;
1286                 if (pm->show_pfn)
1287                         frame = pte_pfn(pte) +
1288                                 ((addr & ~hmask) >> PAGE_SHIFT);
1289         }
1290
1291         for (; addr != end; addr += PAGE_SIZE) {
1292                 pagemap_entry_t pme = make_pme(frame, flags);
1293
1294                 err = add_to_pagemap(addr, &pme, pm);
1295                 if (err)
1296                         return err;
1297                 if (pm->show_pfn && (flags & PM_PRESENT))
1298                         frame++;
1299         }
1300
1301         cond_resched();
1302
1303         return err;
1304 }
1305 #endif /* HUGETLB_PAGE */
1306
1307 /*
1308  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1309  *
1310  * For each page in the address space, this file contains one 64-bit entry
1311  * consisting of the following:
1312  *
1313  * Bits 0-54  page frame number (PFN) if present
1314  * Bits 0-4   swap type if swapped
1315  * Bits 5-54  swap offset if swapped
1316  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1317  * Bit  56    page exclusively mapped
1318  * Bits 57-60 zero
1319  * Bit  61    page is file-page or shared-anon
1320  * Bit  62    page swapped
1321  * Bit  63    page present
1322  *
1323  * If the page is not present but in swap, then the PFN contains an
1324  * encoding of the swap file number and the page's offset into the
1325  * swap. Unmapped pages return a null PFN. This allows determining
1326  * precisely which pages are mapped (or in swap) and comparing mapped
1327  * pages between processes.
1328  *
1329  * Efficient users of this interface will use /proc/pid/maps to
1330  * determine which areas of memory are actually mapped and llseek to
1331  * skip over unmapped regions.
1332  */
1333 static ssize_t pagemap_read(struct file *file, char __user *buf,
1334                             size_t count, loff_t *ppos)
1335 {
1336         struct mm_struct *mm = file->private_data;
1337         struct pagemapread pm;
1338         struct mm_walk pagemap_walk = {};
1339         unsigned long src;
1340         unsigned long svpfn;
1341         unsigned long start_vaddr;
1342         unsigned long end_vaddr;
1343         int ret = 0, copied = 0;
1344
1345         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1346                 goto out;
1347
1348         ret = -EINVAL;
1349         /* file position must be aligned */
1350         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1351                 goto out_mm;
1352
1353         ret = 0;
1354         if (!count)
1355                 goto out_mm;
1356
1357         /* do not disclose physical addresses: attack vector */
1358         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1359
1360         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1361         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1362         ret = -ENOMEM;
1363         if (!pm.buffer)
1364                 goto out_mm;
1365
1366         pagemap_walk.pmd_entry = pagemap_pmd_range;
1367         pagemap_walk.pte_hole = pagemap_pte_hole;
1368 #ifdef CONFIG_HUGETLB_PAGE
1369         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1370 #endif
1371         pagemap_walk.mm = mm;
1372         pagemap_walk.private = &pm;
1373
1374         src = *ppos;
1375         svpfn = src / PM_ENTRY_BYTES;
1376         start_vaddr = svpfn << PAGE_SHIFT;
1377         end_vaddr = mm->task_size;
1378
1379         /* watch out for wraparound */
1380         if (svpfn > mm->task_size >> PAGE_SHIFT)
1381                 start_vaddr = end_vaddr;
1382
1383         /*
1384          * The odds are that this will stop walking way
1385          * before end_vaddr, because the length of the
1386          * user buffer is tracked in "pm", and the walk
1387          * will stop when we hit the end of the buffer.
1388          */
1389         ret = 0;
1390         while (count && (start_vaddr < end_vaddr)) {
1391                 int len;
1392                 unsigned long end;
1393
1394                 pm.pos = 0;
1395                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1396                 /* overflow ? */
1397                 if (end < start_vaddr || end > end_vaddr)
1398                         end = end_vaddr;
1399                 down_read(&mm->mmap_sem);
1400                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1401                 up_read(&mm->mmap_sem);
1402                 start_vaddr = end;
1403
1404                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1405                 if (copy_to_user(buf, pm.buffer, len)) {
1406                         ret = -EFAULT;
1407                         goto out_free;
1408                 }
1409                 copied += len;
1410                 buf += len;
1411                 count -= len;
1412         }
1413         *ppos += copied;
1414         if (!ret || ret == PM_END_OF_BUFFER)
1415                 ret = copied;
1416
1417 out_free:
1418         kfree(pm.buffer);
1419 out_mm:
1420         mmput(mm);
1421 out:
1422         return ret;
1423 }
1424
1425 static int pagemap_open(struct inode *inode, struct file *file)
1426 {
1427         struct mm_struct *mm;
1428
1429         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1430         if (IS_ERR(mm))
1431                 return PTR_ERR(mm);
1432         file->private_data = mm;
1433         return 0;
1434 }
1435
1436 static int pagemap_release(struct inode *inode, struct file *file)
1437 {
1438         struct mm_struct *mm = file->private_data;
1439
1440         if (mm)
1441                 mmdrop(mm);
1442         return 0;
1443 }
1444
1445 const struct file_operations proc_pagemap_operations = {
1446         .llseek         = mem_lseek, /* borrow this */
1447         .read           = pagemap_read,
1448         .open           = pagemap_open,
1449         .release        = pagemap_release,
1450 };
1451 #endif /* CONFIG_PROC_PAGE_MONITOR */
1452
1453 #ifdef CONFIG_NUMA
1454
1455 struct numa_maps {
1456         unsigned long pages;
1457         unsigned long anon;
1458         unsigned long active;
1459         unsigned long writeback;
1460         unsigned long mapcount_max;
1461         unsigned long dirty;
1462         unsigned long swapcache;
1463         unsigned long node[MAX_NUMNODES];
1464 };
1465
1466 struct numa_maps_private {
1467         struct proc_maps_private proc_maps;
1468         struct numa_maps md;
1469 };
1470
1471 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1472                         unsigned long nr_pages)
1473 {
1474         int count = page_mapcount(page);
1475
1476         md->pages += nr_pages;
1477         if (pte_dirty || PageDirty(page))
1478                 md->dirty += nr_pages;
1479
1480         if (PageSwapCache(page))
1481                 md->swapcache += nr_pages;
1482
1483         if (PageActive(page) || PageUnevictable(page))
1484                 md->active += nr_pages;
1485
1486         if (PageWriteback(page))
1487                 md->writeback += nr_pages;
1488
1489         if (PageAnon(page))
1490                 md->anon += nr_pages;
1491
1492         if (count > md->mapcount_max)
1493                 md->mapcount_max = count;
1494
1495         md->node[page_to_nid(page)] += nr_pages;
1496 }
1497
1498 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1499                 unsigned long addr)
1500 {
1501         struct page *page;
1502         int nid;
1503
1504         if (!pte_present(pte))
1505                 return NULL;
1506
1507         page = vm_normal_page(vma, addr, pte);
1508         if (!page)
1509                 return NULL;
1510
1511         if (PageReserved(page))
1512                 return NULL;
1513
1514         nid = page_to_nid(page);
1515         if (!node_isset(nid, node_states[N_MEMORY]))
1516                 return NULL;
1517
1518         return page;
1519 }
1520
1521 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1522                 unsigned long end, struct mm_walk *walk)
1523 {
1524         struct numa_maps *md = walk->private;
1525         struct vm_area_struct *vma = walk->vma;
1526         spinlock_t *ptl;
1527         pte_t *orig_pte;
1528         pte_t *pte;
1529
1530         ptl = pmd_trans_huge_lock(pmd, vma);
1531         if (ptl) {
1532                 pte_t huge_pte = *(pte_t *)pmd;
1533                 struct page *page;
1534
1535                 page = can_gather_numa_stats(huge_pte, vma, addr);
1536                 if (page)
1537                         gather_stats(page, md, pte_dirty(huge_pte),
1538                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1539                 spin_unlock(ptl);
1540                 return 0;
1541         }
1542
1543         if (pmd_trans_unstable(pmd))
1544                 return 0;
1545         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1546         do {
1547                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1548                 if (!page)
1549                         continue;
1550                 gather_stats(page, md, pte_dirty(*pte), 1);
1551
1552         } while (pte++, addr += PAGE_SIZE, addr != end);
1553         pte_unmap_unlock(orig_pte, ptl);
1554         return 0;
1555 }
1556 #ifdef CONFIG_HUGETLB_PAGE
1557 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1558                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1559 {
1560         pte_t huge_pte = huge_ptep_get(pte);
1561         struct numa_maps *md;
1562         struct page *page;
1563
1564         if (!pte_present(huge_pte))
1565                 return 0;
1566
1567         page = pte_page(huge_pte);
1568         if (!page)
1569                 return 0;
1570
1571         md = walk->private;
1572         gather_stats(page, md, pte_dirty(huge_pte), 1);
1573         return 0;
1574 }
1575
1576 #else
1577 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1578                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1579 {
1580         return 0;
1581 }
1582 #endif
1583
1584 /*
1585  * Display pages allocated per node and memory policy via /proc.
1586  */
1587 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1588 {
1589         struct numa_maps_private *numa_priv = m->private;
1590         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1591         struct vm_area_struct *vma = v;
1592         struct numa_maps *md = &numa_priv->md;
1593         struct file *file = vma->vm_file;
1594         struct mm_struct *mm = vma->vm_mm;
1595         struct mm_walk walk = {
1596                 .hugetlb_entry = gather_hugetlb_stats,
1597                 .pmd_entry = gather_pte_stats,
1598                 .private = md,
1599                 .mm = mm,
1600         };
1601         struct mempolicy *pol;
1602         char buffer[64];
1603         int nid;
1604
1605         if (!mm)
1606                 return 0;
1607
1608         /* Ensure we start with an empty set of numa_maps statistics. */
1609         memset(md, 0, sizeof(*md));
1610
1611         pol = __get_vma_policy(vma, vma->vm_start);
1612         if (pol) {
1613                 mpol_to_str(buffer, sizeof(buffer), pol);
1614                 mpol_cond_put(pol);
1615         } else {
1616                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1617         }
1618
1619         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1620
1621         if (file) {
1622                 seq_puts(m, " file=");
1623                 seq_file_path(m, file, "\n\t= ");
1624         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1625                 seq_puts(m, " heap");
1626         } else if (is_stack(proc_priv, vma, is_pid)) {
1627                 seq_puts(m, " stack");
1628         }
1629
1630         if (is_vm_hugetlb_page(vma))
1631                 seq_puts(m, " huge");
1632
1633         /* mmap_sem is held by m_start */
1634         walk_page_vma(vma, &walk);
1635
1636         if (!md->pages)
1637                 goto out;
1638
1639         if (md->anon)
1640                 seq_printf(m, " anon=%lu", md->anon);
1641
1642         if (md->dirty)
1643                 seq_printf(m, " dirty=%lu", md->dirty);
1644
1645         if (md->pages != md->anon && md->pages != md->dirty)
1646                 seq_printf(m, " mapped=%lu", md->pages);
1647
1648         if (md->mapcount_max > 1)
1649                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1650
1651         if (md->swapcache)
1652                 seq_printf(m, " swapcache=%lu", md->swapcache);
1653
1654         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1655                 seq_printf(m, " active=%lu", md->active);
1656
1657         if (md->writeback)
1658                 seq_printf(m, " writeback=%lu", md->writeback);
1659
1660         for_each_node_state(nid, N_MEMORY)
1661                 if (md->node[nid])
1662                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1663
1664         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1665 out:
1666         seq_putc(m, '\n');
1667         m_cache_vma(m, vma);
1668         return 0;
1669 }
1670
1671 static int show_pid_numa_map(struct seq_file *m, void *v)
1672 {
1673         return show_numa_map(m, v, 1);
1674 }
1675
1676 static int show_tid_numa_map(struct seq_file *m, void *v)
1677 {
1678         return show_numa_map(m, v, 0);
1679 }
1680
1681 static const struct seq_operations proc_pid_numa_maps_op = {
1682         .start  = m_start,
1683         .next   = m_next,
1684         .stop   = m_stop,
1685         .show   = show_pid_numa_map,
1686 };
1687
1688 static const struct seq_operations proc_tid_numa_maps_op = {
1689         .start  = m_start,
1690         .next   = m_next,
1691         .stop   = m_stop,
1692         .show   = show_tid_numa_map,
1693 };
1694
1695 static int numa_maps_open(struct inode *inode, struct file *file,
1696                           const struct seq_operations *ops)
1697 {
1698         return proc_maps_open(inode, file, ops,
1699                                 sizeof(struct numa_maps_private));
1700 }
1701
1702 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1703 {
1704         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1705 }
1706
1707 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1708 {
1709         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1710 }
1711
1712 const struct file_operations proc_pid_numa_maps_operations = {
1713         .open           = pid_numa_maps_open,
1714         .read           = seq_read,
1715         .llseek         = seq_lseek,
1716         .release        = proc_map_release,
1717 };
1718
1719 const struct file_operations proc_tid_numa_maps_operations = {
1720         .open           = tid_numa_maps_open,
1721         .read           = seq_read,
1722         .llseek         = seq_lseek,
1723         .release        = proc_map_release,
1724 };
1725 #endif /* CONFIG_NUMA */