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