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