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 if (is_zone_device_page(page))
 585                 /* pass */;
 586         else
 587                 VM_BUG_ON_PAGE(1, page);
 588         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
 589 }
 590 #else
 591 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 592                 struct mm_walk *walk)
 593 {
 594 }
 595 #endif
 596
 597 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 598                            struct mm_walk *walk)
 599 {
 600         struct vm_area_struct *vma = walk->vma;
 601         pte_t *pte;
 602         spinlock_t *ptl;
 603
 604         ptl = pmd_trans_huge_lock(pmd, vma);
 605         if (ptl) {
 606                 smaps_pmd_entry(pmd, addr, walk);
 607                 spin_unlock(ptl);
 608                 return 0;
 609         }
 610
 611         if (pmd_trans_unstable(pmd))
 612                 return 0;
 613         /*
 614          * The mmap_sem held all the way back in m_start() is what
 615          * keeps khugepaged out of here and from collapsing things
 616          * in here.
 617          */
 618         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 619         for (; addr != end; pte++, addr += PAGE_SIZE)
 620                 smaps_pte_entry(pte, addr, walk);
 621         pte_unmap_unlock(pte - 1, ptl);
 622         cond_resched();
 623         return 0;
 624 }
 625
 626 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 627 {
 628         /*
 629          * Don't forget to update Documentation/ on changes.
 630          */
 631         static const char mnemonics[BITS_PER_LONG][2] = {
 632                 /*
 633                  * In case if we meet a flag we don't know about.
 634                  */
 635                 [0 ... (BITS_PER_LONG-1)] = "??",
 636
 637                 [ilog2(VM_READ)]        = "rd",
 638                 [ilog2(VM_WRITE)]       = "wr",
 639                 [ilog2(VM_EXEC)]        = "ex",
 640                 [ilog2(VM_SHARED)]      = "sh",
 641                 [ilog2(VM_MAYREAD)]     = "mr",
 642                 [ilog2(VM_MAYWRITE)]    = "mw",
 643                 [ilog2(VM_MAYEXEC)]     = "me",
 644                 [ilog2(VM_MAYSHARE)]    = "ms",
 645                 [ilog2(VM_GROWSDOWN)]   = "gd",
 646                 [ilog2(VM_PFNMAP)]      = "pf",
 647                 [ilog2(VM_DENYWRITE)]   = "dw",
 648 #ifdef CONFIG_X86_INTEL_MPX
 649                 [ilog2(VM_MPX)]         = "mp",
 650 #endif
 651                 [ilog2(VM_LOCKED)]      = "lo",
 652                 [ilog2(VM_IO)]          = "io",
 653                 [ilog2(VM_SEQ_READ)]    = "sr",
 654                 [ilog2(VM_RAND_READ)]   = "rr",
 655                 [ilog2(VM_DONTCOPY)]    = "dc",
 656                 [ilog2(VM_DONTEXPAND)]  = "de",
 657                 [ilog2(VM_ACCOUNT)]     = "ac",
 658                 [ilog2(VM_NORESERVE)]   = "nr",
 659                 [ilog2(VM_HUGETLB)]     = "ht",
 660                 [ilog2(VM_ARCH_1)]      = "ar",
 661                 [ilog2(VM_DONTDUMP)]    = "dd",
 662 #ifdef CONFIG_MEM_SOFT_DIRTY
 663                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 664 #endif
 665                 [ilog2(VM_MIXEDMAP)]    = "mm",
 666                 [ilog2(VM_HUGEPAGE)]    = "hg",
 667                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 668                 [ilog2(VM_MERGEABLE)]   = "mg",
 669                 [ilog2(VM_UFFD_MISSING)]= "um",
 670                 [ilog2(VM_UFFD_WP)]     = "uw",
 671 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 672                 /* These come out via ProtectionKey: */
 673                 [ilog2(VM_PKEY_BIT0)]   = "",
 674                 [ilog2(VM_PKEY_BIT1)]   = "",
 675                 [ilog2(VM_PKEY_BIT2)]   = "",
 676                 [ilog2(VM_PKEY_BIT3)]   = "",
 677 #endif
 678         };
 679         size_t i;
 680
 681         seq_puts(m, "VmFlags: ");
 682         for (i = 0; i < BITS_PER_LONG; i++) {
 683                 if (!mnemonics[i][0])
 684                         continue;
 685                 if (vma->vm_flags & (1UL << i)) {
 686                         seq_printf(m, "%c%c ",
 687                                    mnemonics[i][0], mnemonics[i][1]);
 688                 }
 689         }
 690         seq_putc(m, '\n');
 691 }
 692
 693 #ifdef CONFIG_HUGETLB_PAGE
 694 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 695                                  unsigned long addr, unsigned long end,
 696                                  struct mm_walk *walk)
 697 {
 698         struct mem_size_stats *mss = walk->private;
 699         struct vm_area_struct *vma = walk->vma;
 700         struct page *page = NULL;
 701
 702         if (pte_present(*pte)) {
 703                 page = vm_normal_page(vma, addr, *pte);
 704         } else if (is_swap_pte(*pte)) {
 705                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 706
 707                 if (is_migration_entry(swpent))
 708                         page = migration_entry_to_page(swpent);
 709         }
 710         if (page) {
 711                 int mapcount = page_mapcount(page);
 712
 713                 if (mapcount >= 2)
 714                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 715                 else
 716                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 717         }
 718         return 0;
 719 }
 720 #endif /* HUGETLB_PAGE */
 721
 722 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
 723 {
 724 }
 725
 726 static int show_smap(struct seq_file *m, void *v, int is_pid)
 727 {
 728         struct vm_area_struct *vma = v;
 729         struct mem_size_stats mss;
 730         struct mm_walk smaps_walk = {
 731                 .pmd_entry = smaps_pte_range,
 732 #ifdef CONFIG_HUGETLB_PAGE
 733                 .hugetlb_entry = smaps_hugetlb_range,
 734 #endif
 735                 .mm = vma->vm_mm,
 736                 .private = &mss,
 737         };
 738
 739         memset(&mss, 0, sizeof mss);
 740
 741 #ifdef CONFIG_SHMEM
 742         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 743                 /*
 744                  * For shared or readonly shmem mappings we know that all
 745                  * swapped out pages belong to the shmem object, and we can
 746                  * obtain the swap value much more efficiently. For private
 747                  * writable mappings, we might have COW pages that are
 748                  * not affected by the parent swapped out pages of the shmem
 749                  * object, so we have to distinguish them during the page walk.
 750                  * Unless we know that the shmem object (or the part mapped by
 751                  * our VMA) has no swapped out pages at all.
 752                  */
 753                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 754
 755                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 756                                         !(vma->vm_flags & VM_WRITE)) {
 757                         mss.swap = shmem_swapped;
 758                 } else {
 759                         mss.check_shmem_swap = true;
 760                         smaps_walk.pte_hole = smaps_pte_hole;
 761                 }
 762         }
 763 #endif
 764
 765         /* mmap_sem is held in m_start */
 766         walk_page_vma(vma, &smaps_walk);
 767
 768         show_map_vma(m, vma, is_pid);
 769
 770         seq_printf(m,
 771                    "Size:           %8lu kB\n"
 772                    "Rss:            %8lu kB\n"
 773                    "Pss:            %8lu kB\n"
 774                    "Shared_Clean:   %8lu kB\n"
 775                    "Shared_Dirty:   %8lu kB\n"
 776                    "Private_Clean:  %8lu kB\n"
 777                    "Private_Dirty:  %8lu kB\n"
 778                    "Referenced:     %8lu kB\n"
 779                    "Anonymous:      %8lu kB\n"
 780                    "AnonHugePages:  %8lu kB\n"
 781                    "ShmemPmdMapped: %8lu kB\n"
 782                    "Shared_Hugetlb: %8lu kB\n"
 783                    "Private_Hugetlb: %7lu kB\n"
 784                    "Swap:           %8lu kB\n"
 785                    "SwapPss:        %8lu kB\n"
 786                    "KernelPageSize: %8lu kB\n"
 787                    "MMUPageSize:    %8lu kB\n"
 788                    "Locked:         %8lu kB\n",
 789                    (vma->vm_end - vma->vm_start) >> 10,
 790                    mss.resident >> 10,
 791                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 792                    mss.shared_clean  >> 10,
 793                    mss.shared_dirty  >> 10,
 794                    mss.private_clean >> 10,
 795                    mss.private_dirty >> 10,
 796                    mss.referenced >> 10,
 797                    mss.anonymous >> 10,
 798                    mss.anonymous_thp >> 10,
 799                    mss.shmem_thp >> 10,
 800                    mss.shared_hugetlb >> 10,
 801                    mss.private_hugetlb >> 10,
 802                    mss.swap >> 10,
 803                    (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
 804                    vma_kernel_pagesize(vma) >> 10,
 805                    vma_mmu_pagesize(vma) >> 10,
 806                    (vma->vm_flags & VM_LOCKED) ?
 807                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 808
 809         arch_show_smap(m, vma);
 810         show_smap_vma_flags(m, vma);
 811         m_cache_vma(m, vma);
 812         return 0;
 813 }
 814
 815 static int show_pid_smap(struct seq_file *m, void *v)
 816 {
 817         return show_smap(m, v, 1);
 818 }
 819
 820 static int show_tid_smap(struct seq_file *m, void *v)
 821 {
 822         return show_smap(m, v, 0);
 823 }
 824
 825 static const struct seq_operations proc_pid_smaps_op = {
 826         .start  = m_start,
 827         .next   = m_next,
 828         .stop   = m_stop,
 829         .show   = show_pid_smap
 830 };
 831
 832 static const struct seq_operations proc_tid_smaps_op = {
 833         .start  = m_start,
 834         .next   = m_next,
 835         .stop   = m_stop,
 836         .show   = show_tid_smap
 837 };
 838
 839 static int pid_smaps_open(struct inode *inode, struct file *file)
 840 {
 841         return do_maps_open(inode, file, &proc_pid_smaps_op);
 842 }
 843
 844 static int tid_smaps_open(struct inode *inode, struct file *file)
 845 {
 846         return do_maps_open(inode, file, &proc_tid_smaps_op);
 847 }
 848
 849 const struct file_operations proc_pid_smaps_operations = {
 850         .open           = pid_smaps_open,
 851         .read           = seq_read,
 852         .llseek         = seq_lseek,
 853         .release        = proc_map_release,
 854 };
 855
 856 const struct file_operations proc_tid_smaps_operations = {
 857         .open           = tid_smaps_open,
 858         .read           = seq_read,
 859         .llseek         = seq_lseek,
 860         .release        = proc_map_release,
 861 };
 862
 863 enum clear_refs_types {
 864         CLEAR_REFS_ALL = 1,
 865         CLEAR_REFS_ANON,
 866         CLEAR_REFS_MAPPED,
 867         CLEAR_REFS_SOFT_DIRTY,
 868         CLEAR_REFS_MM_HIWATER_RSS,
 869         CLEAR_REFS_LAST,
 870 };
 871
 872 struct clear_refs_private {
 873         enum clear_refs_types type;
 874 };
 875
 876 #ifdef CONFIG_MEM_SOFT_DIRTY
 877 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 878                 unsigned long addr, pte_t *pte)
 879 {
 880         /*
 881          * The soft-dirty tracker uses #PF-s to catch writes
 882          * to pages, so write-protect the pte as well. See the
 883          * Documentation/vm/soft-dirty.txt for full description
 884          * of how soft-dirty works.
 885          */
 886         pte_t ptent = *pte;
 887
 888         if (pte_present(ptent)) {
 889                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 890                 ptent = pte_wrprotect(ptent);
 891                 ptent = pte_clear_soft_dirty(ptent);
 892                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 893         } else if (is_swap_pte(ptent)) {
 894                 ptent = pte_swp_clear_soft_dirty(ptent);
 895                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 896         }
 897 }
 898 #else
 899 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 900                 unsigned long addr, pte_t *pte)
 901 {
 902 }
 903 #endif
 904
 905 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 906 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 907                 unsigned long addr, pmd_t *pmdp)
 908 {
 909         pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
 910
 911         pmd = pmd_wrprotect(pmd);
 912         pmd = pmd_clear_soft_dirty(pmd);
 913
 914         set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 915 }
 916 #else
 917 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 918                 unsigned long addr, pmd_t *pmdp)
 919 {
 920 }
 921 #endif
 922
 923 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 924                                 unsigned long end, struct mm_walk *walk)
 925 {
 926         struct clear_refs_private *cp = walk->private;
 927         struct vm_area_struct *vma = walk->vma;
 928         pte_t *pte, ptent;
 929         spinlock_t *ptl;
 930         struct page *page;
 931
 932         ptl = pmd_trans_huge_lock(pmd, vma);
 933         if (ptl) {
 934                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 935                         clear_soft_dirty_pmd(vma, addr, pmd);
 936                         goto out;
 937                 }
 938
 939                 page = pmd_page(*pmd);
 940
 941                 /* Clear accessed and referenced bits. */
 942                 pmdp_test_and_clear_young(vma, addr, pmd);
 943                 test_and_clear_page_young(page);
 944                 ClearPageReferenced(page);
 945 out:
 946                 spin_unlock(ptl);
 947                 return 0;
 948         }
 949
 950         if (pmd_trans_unstable(pmd))
 951                 return 0;
 952
 953         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 954         for (; addr != end; pte++, addr += PAGE_SIZE) {
 955                 ptent = *pte;
 956
 957                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 958                         clear_soft_dirty(vma, addr, pte);
 959                         continue;
 960                 }
 961
 962                 if (!pte_present(ptent))
 963                         continue;
 964
 965                 page = vm_normal_page(vma, addr, ptent);
 966                 if (!page)
 967                         continue;
 968
 969                 /* Clear accessed and referenced bits. */
 970                 ptep_test_and_clear_young(vma, addr, pte);
 971                 test_and_clear_page_young(page);
 972                 ClearPageReferenced(page);
 973         }
 974         pte_unmap_unlock(pte - 1, ptl);
 975         cond_resched();
 976         return 0;
 977 }
 978
 979 static int clear_refs_test_walk(unsigned long start, unsigned long end,
 980                                 struct mm_walk *walk)
 981 {
 982         struct clear_refs_private *cp = walk->private;
 983         struct vm_area_struct *vma = walk->vma;
 984
 985         if (vma->vm_flags & VM_PFNMAP)
 986                 return 1;
 987
 988         /*
 989          * Writing 1 to /proc/pid/clear_refs affects all pages.
 990          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
 991          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
 992          * Writing 4 to /proc/pid/clear_refs affects all pages.
 993          */
 994         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
 995                 return 1;
 996         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
 997                 return 1;
 998         return 0;
 999 }
1000
1001 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1002                                 size_t count, loff_t *ppos)
1003 {
1004         struct task_struct *task;
1005         char buffer[PROC_NUMBUF];
1006         struct mm_struct *mm;
1007         struct vm_area_struct *vma;
1008         enum clear_refs_types type;
1009         int itype;
1010         int rv;
1011
1012         memset(buffer, 0, sizeof(buffer));
1013         if (count > sizeof(buffer) - 1)
1014                 count = sizeof(buffer) - 1;
1015         if (copy_from_user(buffer, buf, count))
1016                 return -EFAULT;
1017         rv = kstrtoint(strstrip(buffer), 10, &itype);
1018         if (rv < 0)
1019                 return rv;
1020         type = (enum clear_refs_types)itype;
1021         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1022                 return -EINVAL;
1023
1024         task = get_proc_task(file_inode(file));
1025         if (!task)
1026                 return -ESRCH;
1027         mm = get_task_mm(task);
1028         if (mm) {
1029                 struct clear_refs_private cp = {
1030                         .type = type,
1031                 };
1032                 struct mm_walk clear_refs_walk = {
1033                         .pmd_entry = clear_refs_pte_range,
1034                         .test_walk = clear_refs_test_walk,
1035                         .mm = mm,
1036                         .private = &cp,
1037                 };
1038
1039                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1040                         if (down_write_killable(&mm->mmap_sem)) {
1041                                 count = -EINTR;
1042                                 goto out_mm;
1043                         }
1044
1045                         /*
1046                          * Writing 5 to /proc/pid/clear_refs resets the peak
1047                          * resident set size to this mm's current rss value.
1048                          */
1049                         reset_mm_hiwater_rss(mm);
1050                         up_write(&mm->mmap_sem);
1051                         goto out_mm;
1052                 }
1053
1054                 down_read(&mm->mmap_sem);
1055                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1056                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1057                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1058                                         continue;
1059                                 up_read(&mm->mmap_sem);
1060                                 if (down_write_killable(&mm->mmap_sem)) {
1061                                         count = -EINTR;
1062                                         goto out_mm;
1063                                 }
1064                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1065                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1066                                         vma_set_page_prot(vma);
1067                                 }
1068                                 downgrade_write(&mm->mmap_sem);
1069                                 break;
1070                         }
1071                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1072                 }
1073                 walk_page_range(0, ~0UL, &clear_refs_walk);
1074                 if (type == CLEAR_REFS_SOFT_DIRTY)
1075                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1076                 flush_tlb_mm(mm);
1077                 up_read(&mm->mmap_sem);
1078 out_mm:
1079                 mmput(mm);
1080         }
1081         put_task_struct(task);
1082
1083         return count;
1084 }
1085
1086 const struct file_operations proc_clear_refs_operations = {
1087         .write          = clear_refs_write,
1088         .llseek         = noop_llseek,
1089 };
1090
1091 typedef struct {
1092         u64 pme;
1093 } pagemap_entry_t;
1094
1095 struct pagemapread {
1096         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1097         pagemap_entry_t *buffer;
1098         bool show_pfn;
1099 };
1100
1101 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1102 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1103
1104 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1105 #define PM_PFRAME_BITS          55
1106 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1107 #define PM_SOFT_DIRTY           BIT_ULL(55)
1108 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1109 #define PM_FILE                 BIT_ULL(61)
1110 #define PM_SWAP                 BIT_ULL(62)
1111 #define PM_PRESENT              BIT_ULL(63)
1112
1113 #define PM_END_OF_BUFFER    1
1114
1115 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1116 {
1117         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1118 }
1119
1120 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1121                           struct pagemapread *pm)
1122 {
1123         pm->buffer[pm->pos++] = *pme;
1124         if (pm->pos >= pm->len)
1125                 return PM_END_OF_BUFFER;
1126         return 0;
1127 }
1128
1129 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1130                                 struct mm_walk *walk)
1131 {
1132         struct pagemapread *pm = walk->private;
1133         unsigned long addr = start;
1134         int err = 0;
1135
1136         while (addr < end) {
1137                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1138                 pagemap_entry_t pme = make_pme(0, 0);
1139                 /* End of address space hole, which we mark as non-present. */
1140                 unsigned long hole_end;
1141
1142                 if (vma)
1143                         hole_end = min(end, vma->vm_start);
1144                 else
1145                         hole_end = end;
1146
1147                 for (; addr < hole_end; addr += PAGE_SIZE) {
1148                         err = add_to_pagemap(addr, &pme, pm);
1149                         if (err)
1150                                 goto out;
1151                 }
1152
1153                 if (!vma)
1154                         break;
1155
1156                 /* Addresses in the VMA. */
1157                 if (vma->vm_flags & VM_SOFTDIRTY)
1158                         pme = make_pme(0, PM_SOFT_DIRTY);
1159                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1160                         err = add_to_pagemap(addr, &pme, pm);
1161                         if (err)
1162                                 goto out;
1163                 }
1164         }
1165 out:
1166         return err;
1167 }
1168
1169 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1170                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1171 {
1172         u64 frame = 0, flags = 0;
1173         struct page *page = NULL;
1174
1175         if (pte_present(pte)) {
1176                 if (pm->show_pfn)
1177                         frame = pte_pfn(pte);
1178                 flags |= PM_PRESENT;
1179                 page = vm_normal_page(vma, addr, pte);
1180                 if (pte_soft_dirty(pte))
1181                         flags |= PM_SOFT_DIRTY;
1182         } else if (is_swap_pte(pte)) {
1183                 swp_entry_t entry;
1184                 if (pte_swp_soft_dirty(pte))
1185                         flags |= PM_SOFT_DIRTY;
1186                 entry = pte_to_swp_entry(pte);
1187                 frame = swp_type(entry) |
1188                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1189                 flags |= PM_SWAP;
1190                 if (is_migration_entry(entry))
1191                         page = migration_entry_to_page(entry);
1192         }
1193
1194         if (page && !PageAnon(page))
1195                 flags |= PM_FILE;
1196         if (page && page_mapcount(page) == 1)
1197                 flags |= PM_MMAP_EXCLUSIVE;
1198         if (vma->vm_flags & VM_SOFTDIRTY)
1199                 flags |= PM_SOFT_DIRTY;
1200
1201         return make_pme(frame, flags);
1202 }
1203
1204 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1205                              struct mm_walk *walk)
1206 {
1207         struct vm_area_struct *vma = walk->vma;
1208         struct pagemapread *pm = walk->private;
1209         spinlock_t *ptl;
1210         pte_t *pte, *orig_pte;
1211         int err = 0;
1212
1213 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1214         ptl = pmd_trans_huge_lock(pmdp, vma);
1215         if (ptl) {
1216                 u64 flags = 0, frame = 0;
1217                 pmd_t pmd = *pmdp;
1218
1219                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1220                         flags |= PM_SOFT_DIRTY;
1221
1222                 /*
1223                  * Currently pmd for thp is always present because thp
1224                  * can not be swapped-out, migrated, or HWPOISONed
1225                  * (split in such cases instead.)
1226                  * This if-check is just to prepare for future implementation.
1227                  */
1228                 if (pmd_present(pmd)) {
1229                         struct page *page = pmd_page(pmd);
1230
1231                         if (page_mapcount(page) == 1)
1232                                 flags |= PM_MMAP_EXCLUSIVE;
1233
1234                         flags |= PM_PRESENT;
1235                         if (pm->show_pfn)
1236                                 frame = pmd_pfn(pmd) +
1237                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1238                 }
1239
1240                 for (; addr != end; addr += PAGE_SIZE) {
1241                         pagemap_entry_t pme = make_pme(frame, flags);
1242
1243                         err = add_to_pagemap(addr, &pme, pm);
1244                         if (err)
1245                                 break;
1246                         if (pm->show_pfn && (flags & PM_PRESENT))
1247                                 frame++;
1248                 }
1249                 spin_unlock(ptl);
1250                 return err;
1251         }
1252
1253         if (pmd_trans_unstable(pmdp))
1254                 return 0;
1255 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1256
1257         /*
1258          * We can assume that @vma always points to a valid one and @end never
1259          * goes beyond vma->vm_end.
1260          */
1261         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1262         for (; addr < end; pte++, addr += PAGE_SIZE) {
1263                 pagemap_entry_t pme;
1264
1265                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1266                 err = add_to_pagemap(addr, &pme, pm);
1267                 if (err)
1268                         break;
1269         }
1270         pte_unmap_unlock(orig_pte, ptl);
1271
1272         cond_resched();
1273
1274         return err;
1275 }
1276
1277 #ifdef CONFIG_HUGETLB_PAGE
1278 /* This function walks within one hugetlb entry in the single call */
1279 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1280                                  unsigned long addr, unsigned long end,
1281                                  struct mm_walk *walk)
1282 {
1283         struct pagemapread *pm = walk->private;
1284         struct vm_area_struct *vma = walk->vma;
1285         u64 flags = 0, frame = 0;
1286         int err = 0;
1287         pte_t pte;
1288
1289         if (vma->vm_flags & VM_SOFTDIRTY)
1290                 flags |= PM_SOFT_DIRTY;
1291
1292         pte = huge_ptep_get(ptep);
1293         if (pte_present(pte)) {
1294                 struct page *page = pte_page(pte);
1295
1296                 if (!PageAnon(page))
1297                         flags |= PM_FILE;
1298
1299                 if (page_mapcount(page) == 1)
1300                         flags |= PM_MMAP_EXCLUSIVE;
1301
1302                 flags |= PM_PRESENT;
1303                 if (pm->show_pfn)
1304                         frame = pte_pfn(pte) +
1305                                 ((addr & ~hmask) >> PAGE_SHIFT);
1306         }
1307
1308         for (; addr != end; addr += PAGE_SIZE) {
1309                 pagemap_entry_t pme = make_pme(frame, flags);
1310
1311                 err = add_to_pagemap(addr, &pme, pm);
1312                 if (err)
1313                         return err;
1314                 if (pm->show_pfn && (flags & PM_PRESENT))
1315                         frame++;
1316         }
1317
1318         cond_resched();
1319
1320         return err;
1321 }
1322 #endif /* HUGETLB_PAGE */
1323
1324 /*
1325  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1326  *
1327  * For each page in the address space, this file contains one 64-bit entry
1328  * consisting of the following:
1329  *
1330  * Bits 0-54  page frame number (PFN) if present
1331  * Bits 0-4   swap type if swapped
1332  * Bits 5-54  swap offset if swapped
1333  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1334  * Bit  56    page exclusively mapped
1335  * Bits 57-60 zero
1336  * Bit  61    page is file-page or shared-anon
1337  * Bit  62    page swapped
1338  * Bit  63    page present
1339  *
1340  * If the page is not present but in swap, then the PFN contains an
1341  * encoding of the swap file number and the page's offset into the
1342  * swap. Unmapped pages return a null PFN. This allows determining
1343  * precisely which pages are mapped (or in swap) and comparing mapped
1344  * pages between processes.
1345  *
1346  * Efficient users of this interface will use /proc/pid/maps to
1347  * determine which areas of memory are actually mapped and llseek to
1348  * skip over unmapped regions.
1349  */
1350 static ssize_t pagemap_read(struct file *file, char __user *buf,
1351                             size_t count, loff_t *ppos)
1352 {
1353         struct mm_struct *mm = file->private_data;
1354         struct pagemapread pm;
1355         struct mm_walk pagemap_walk = {};
1356         unsigned long src;
1357         unsigned long svpfn;
1358         unsigned long start_vaddr;
1359         unsigned long end_vaddr;
1360         int ret = 0, copied = 0;
1361
1362         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1363                 goto out;
1364
1365         ret = -EINVAL;
1366         /* file position must be aligned */
1367         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1368                 goto out_mm;
1369
1370         ret = 0;
1371         if (!count)
1372                 goto out_mm;
1373
1374         /* do not disclose physical addresses: attack vector */
1375         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1376
1377         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1378         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1379         ret = -ENOMEM;
1380         if (!pm.buffer)
1381                 goto out_mm;
1382
1383         pagemap_walk.pmd_entry = pagemap_pmd_range;
1384         pagemap_walk.pte_hole = pagemap_pte_hole;
1385 #ifdef CONFIG_HUGETLB_PAGE
1386         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1387 #endif
1388         pagemap_walk.mm = mm;
1389         pagemap_walk.private = &pm;
1390
1391         src = *ppos;
1392         svpfn = src / PM_ENTRY_BYTES;
1393         start_vaddr = svpfn << PAGE_SHIFT;
1394         end_vaddr = mm->task_size;
1395
1396         /* watch out for wraparound */
1397         if (svpfn > mm->task_size >> PAGE_SHIFT)
1398                 start_vaddr = end_vaddr;
1399
1400         /*
1401          * The odds are that this will stop walking way
1402          * before end_vaddr, because the length of the
1403          * user buffer is tracked in "pm", and the walk
1404          * will stop when we hit the end of the buffer.
1405          */
1406         ret = 0;
1407         while (count && (start_vaddr < end_vaddr)) {
1408                 int len;
1409                 unsigned long end;
1410
1411                 pm.pos = 0;
1412                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1413                 /* overflow ? */
1414                 if (end < start_vaddr || end > end_vaddr)
1415                         end = end_vaddr;
1416                 down_read(&mm->mmap_sem);
1417                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1418                 up_read(&mm->mmap_sem);
1419                 start_vaddr = end;
1420
1421                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1422                 if (copy_to_user(buf, pm.buffer, len)) {
1423                         ret = -EFAULT;
1424                         goto out_free;
1425                 }
1426                 copied += len;
1427                 buf += len;
1428                 count -= len;
1429         }
1430         *ppos += copied;
1431         if (!ret || ret == PM_END_OF_BUFFER)
1432                 ret = copied;
1433
1434 out_free:
1435         kfree(pm.buffer);
1436 out_mm:
1437         mmput(mm);
1438 out:
1439         return ret;
1440 }
1441
1442 static int pagemap_open(struct inode *inode, struct file *file)
1443 {
1444         struct mm_struct *mm;
1445
1446         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1447         if (IS_ERR(mm))
1448                 return PTR_ERR(mm);
1449         file->private_data = mm;
1450         return 0;
1451 }
1452
1453 static int pagemap_release(struct inode *inode, struct file *file)
1454 {
1455         struct mm_struct *mm = file->private_data;
1456
1457         if (mm)
1458                 mmdrop(mm);
1459         return 0;
1460 }
1461
1462 const struct file_operations proc_pagemap_operations = {
1463         .llseek         = mem_lseek, /* borrow this */
1464         .read           = pagemap_read,
1465         .open           = pagemap_open,
1466         .release        = pagemap_release,
1467 };
1468 #endif /* CONFIG_PROC_PAGE_MONITOR */
1469
1470 #ifdef CONFIG_NUMA
1471
1472 struct numa_maps {
1473         unsigned long pages;
1474         unsigned long anon;
1475         unsigned long active;
1476         unsigned long writeback;
1477         unsigned long mapcount_max;
1478         unsigned long dirty;
1479         unsigned long swapcache;
1480         unsigned long node[MAX_NUMNODES];
1481 };
1482
1483 struct numa_maps_private {
1484         struct proc_maps_private proc_maps;
1485         struct numa_maps md;
1486 };
1487
1488 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1489                         unsigned long nr_pages)
1490 {
1491         int count = page_mapcount(page);
1492
1493         md->pages += nr_pages;
1494         if (pte_dirty || PageDirty(page))
1495                 md->dirty += nr_pages;
1496
1497         if (PageSwapCache(page))
1498                 md->swapcache += nr_pages;
1499
1500         if (PageActive(page) || PageUnevictable(page))
1501                 md->active += nr_pages;
1502
1503         if (PageWriteback(page))
1504                 md->writeback += nr_pages;
1505
1506         if (PageAnon(page))
1507                 md->anon += nr_pages;
1508
1509         if (count > md->mapcount_max)
1510                 md->mapcount_max = count;
1511
1512         md->node[page_to_nid(page)] += nr_pages;
1513 }
1514
1515 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1516                 unsigned long addr)
1517 {
1518         struct page *page;
1519         int nid;
1520
1521         if (!pte_present(pte))
1522                 return NULL;
1523
1524         page = vm_normal_page(vma, addr, pte);
1525         if (!page)
1526                 return NULL;
1527
1528         if (PageReserved(page))
1529                 return NULL;
1530
1531         nid = page_to_nid(page);
1532         if (!node_isset(nid, node_states[N_MEMORY]))
1533                 return NULL;
1534
1535         return page;
1536 }
1537
1538 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1539 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1540                                               struct vm_area_struct *vma,
1541                                               unsigned long addr)
1542 {
1543         struct page *page;
1544         int nid;
1545
1546         if (!pmd_present(pmd))
1547                 return NULL;
1548
1549         page = vm_normal_page_pmd(vma, addr, pmd);
1550         if (!page)
1551                 return NULL;
1552
1553         if (PageReserved(page))
1554                 return NULL;
1555
1556         nid = page_to_nid(page);
1557         if (!node_isset(nid, node_states[N_MEMORY]))
1558                 return NULL;
1559
1560         return page;
1561 }
1562 #endif
1563
1564 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1565                 unsigned long end, struct mm_walk *walk)
1566 {
1567         struct numa_maps *md = walk->private;
1568         struct vm_area_struct *vma = walk->vma;
1569         spinlock_t *ptl;
1570         pte_t *orig_pte;
1571         pte_t *pte;
1572
1573 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1574         ptl = pmd_trans_huge_lock(pmd, vma);
1575         if (ptl) {
1576                 struct page *page;
1577
1578                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1579                 if (page)
1580                         gather_stats(page, md, pmd_dirty(*pmd),
1581                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1582                 spin_unlock(ptl);
1583                 return 0;
1584         }
1585
1586         if (pmd_trans_unstable(pmd))
1587                 return 0;
1588 #endif
1589         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1590         do {
1591                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1592                 if (!page)
1593                         continue;
1594                 gather_stats(page, md, pte_dirty(*pte), 1);
1595
1596         } while (pte++, addr += PAGE_SIZE, addr != end);
1597         pte_unmap_unlock(orig_pte, ptl);
1598         return 0;
1599 }
1600 #ifdef CONFIG_HUGETLB_PAGE
1601 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1602                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1603 {
1604         pte_t huge_pte = huge_ptep_get(pte);
1605         struct numa_maps *md;
1606         struct page *page;
1607
1608         if (!pte_present(huge_pte))
1609                 return 0;
1610
1611         page = pte_page(huge_pte);
1612         if (!page)
1613                 return 0;
1614
1615         md = walk->private;
1616         gather_stats(page, md, pte_dirty(huge_pte), 1);
1617         return 0;
1618 }
1619
1620 #else
1621 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1622                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1623 {
1624         return 0;
1625 }
1626 #endif
1627
1628 /*
1629  * Display pages allocated per node and memory policy via /proc.
1630  */
1631 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1632 {
1633         struct numa_maps_private *numa_priv = m->private;
1634         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1635         struct vm_area_struct *vma = v;
1636         struct numa_maps *md = &numa_priv->md;
1637         struct file *file = vma->vm_file;
1638         struct mm_struct *mm = vma->vm_mm;
1639         struct mm_walk walk = {
1640                 .hugetlb_entry = gather_hugetlb_stats,
1641                 .pmd_entry = gather_pte_stats,
1642                 .private = md,
1643                 .mm = mm,
1644         };
1645         struct mempolicy *pol;
1646         char buffer[64];
1647         int nid;
1648
1649         if (!mm)
1650                 return 0;
1651
1652         /* Ensure we start with an empty set of numa_maps statistics. */
1653         memset(md, 0, sizeof(*md));
1654
1655         pol = __get_vma_policy(vma, vma->vm_start);
1656         if (pol) {
1657                 mpol_to_str(buffer, sizeof(buffer), pol);
1658                 mpol_cond_put(pol);
1659         } else {
1660                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1661         }
1662
1663         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1664
1665         if (file) {
1666                 seq_puts(m, " file=");
1667                 seq_file_path(m, file, "\n\t= ");
1668         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1669                 seq_puts(m, " heap");
1670         } else if (is_stack(proc_priv, vma, is_pid)) {
1671                 seq_puts(m, " stack");
1672         }
1673
1674         if (is_vm_hugetlb_page(vma))
1675                 seq_puts(m, " huge");
1676
1677         /* mmap_sem is held by m_start */
1678         walk_page_vma(vma, &walk);
1679
1680         if (!md->pages)
1681                 goto out;
1682
1683         if (md->anon)
1684                 seq_printf(m, " anon=%lu", md->anon);
1685
1686         if (md->dirty)
1687                 seq_printf(m, " dirty=%lu", md->dirty);
1688
1689         if (md->pages != md->anon && md->pages != md->dirty)
1690                 seq_printf(m, " mapped=%lu", md->pages);
1691
1692         if (md->mapcount_max > 1)
1693                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1694
1695         if (md->swapcache)
1696                 seq_printf(m, " swapcache=%lu", md->swapcache);
1697
1698         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1699                 seq_printf(m, " active=%lu", md->active);
1700
1701         if (md->writeback)
1702                 seq_printf(m, " writeback=%lu", md->writeback);
1703
1704         for_each_node_state(nid, N_MEMORY)
1705                 if (md->node[nid])
1706                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1707
1708         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1709 out:
1710         seq_putc(m, '\n');
1711         m_cache_vma(m, vma);
1712         return 0;
1713 }
1714
1715 static int show_pid_numa_map(struct seq_file *m, void *v)
1716 {
1717         return show_numa_map(m, v, 1);
1718 }
1719
1720 static int show_tid_numa_map(struct seq_file *m, void *v)
1721 {
1722         return show_numa_map(m, v, 0);
1723 }
1724
1725 static const struct seq_operations proc_pid_numa_maps_op = {
1726         .start  = m_start,
1727         .next   = m_next,
1728         .stop   = m_stop,
1729         .show   = show_pid_numa_map,
1730 };
1731
1732 static const struct seq_operations proc_tid_numa_maps_op = {
1733         .start  = m_start,
1734         .next   = m_next,
1735         .stop   = m_stop,
1736         .show   = show_tid_numa_map,
1737 };
1738
1739 static int numa_maps_open(struct inode *inode, struct file *file,
1740                           const struct seq_operations *ops)
1741 {
1742         return proc_maps_open(inode, file, ops,
1743                                 sizeof(struct numa_maps_private));
1744 }
1745
1746 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1747 {
1748         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1749 }
1750
1751 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1752 {
1753         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1754 }
1755
1756 const struct file_operations proc_pid_numa_maps_operations = {
1757         .open           = pid_numa_maps_open,
1758         .read           = seq_read,
1759         .llseek         = seq_lseek,
1760         .release        = proc_map_release,
1761 };
1762
1763 const struct file_operations proc_tid_numa_maps_operations = {
1764         .open           = tid_numa_maps_open,
1765         .read           = seq_read,
1766         .llseek         = seq_lseek,
1767         .release        = proc_map_release,
1768 };
1769 #endif /* CONFIG_NUMA */