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