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
  18 #include <asm/elf.h>
  19 #include <asm/uaccess.h>
  20 #include <asm/tlbflush.h>
  21 #include "internal.h"
  22
  23 void task_mem(struct seq_file *m, struct mm_struct *mm)
  24 {
  25         unsigned long data, text, lib, swap, ptes, pmds;
  26         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  27
  28         /*
  29          * Note: to minimize their overhead, mm maintains hiwater_vm and
  30          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  31          * collector of these hiwater stats must therefore get total_vm
  32          * and rss too, which will usually be the higher.  Barriers? not
  33          * worth the effort, such snapshots can always be inconsistent.
  34          */
  35         hiwater_vm = total_vm = mm->total_vm;
  36         if (hiwater_vm < mm->hiwater_vm)
  37                 hiwater_vm = mm->hiwater_vm;
  38         hiwater_rss = total_rss = get_mm_rss(mm);
  39         if (hiwater_rss < mm->hiwater_rss)
  40                 hiwater_rss = mm->hiwater_rss;
  41
  42         data = mm->total_vm - mm->shared_vm - mm->stack_vm;
  43         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  44         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  45         swap = get_mm_counter(mm, MM_SWAPENTS);
  46         ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
  47         pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
  48         seq_printf(m,
  49                 "VmPeak:\t%8lu kB\n"
  50                 "VmSize:\t%8lu kB\n"
  51                 "VmLck:\t%8lu kB\n"
  52                 "VmPin:\t%8lu kB\n"
  53                 "VmHWM:\t%8lu kB\n"
  54                 "VmRSS:\t%8lu kB\n"
  55                 "VmData:\t%8lu kB\n"
  56                 "VmStk:\t%8lu kB\n"
  57                 "VmExe:\t%8lu kB\n"
  58                 "VmLib:\t%8lu kB\n"
  59                 "VmPTE:\t%8lu kB\n"
  60                 "VmPMD:\t%8lu kB\n"
  61                 "VmSwap:\t%8lu kB\n",
  62                 hiwater_vm << (PAGE_SHIFT-10),
  63                 total_vm << (PAGE_SHIFT-10),
  64                 mm->locked_vm << (PAGE_SHIFT-10),
  65                 mm->pinned_vm << (PAGE_SHIFT-10),
  66                 hiwater_rss << (PAGE_SHIFT-10),
  67                 total_rss << (PAGE_SHIFT-10),
  68                 data << (PAGE_SHIFT-10),
  69                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  70                 ptes >> 10,
  71                 pmds >> 10,
  72                 swap << (PAGE_SHIFT-10));
  73         hugetlb_report_usage(m, mm);
  74 }
  75
  76 unsigned long task_vsize(struct mm_struct *mm)
  77 {
  78         return PAGE_SIZE * mm->total_vm;
  79 }
  80
  81 unsigned long task_statm(struct mm_struct *mm,
  82                          unsigned long *shared, unsigned long *text,
  83                          unsigned long *data, unsigned long *resident)
  84 {
  85         *shared = get_mm_counter(mm, MM_FILEPAGES);
  86         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  87                                                                 >> PAGE_SHIFT;
  88         *data = mm->total_vm - mm->shared_vm;
  89         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
  90         return mm->total_vm;
  91 }
  92
  93 #ifdef CONFIG_NUMA
  94 /*
  95  * Save get_task_policy() for show_numa_map().
  96  */
  97 static void hold_task_mempolicy(struct proc_maps_private *priv)
  98 {
  99         struct task_struct *task = priv->task;
 100
 101         task_lock(task);
 102         priv->task_mempolicy = get_task_policy(task);
 103         mpol_get(priv->task_mempolicy);
 104         task_unlock(task);
 105 }
 106 static void release_task_mempolicy(struct proc_maps_private *priv)
 107 {
 108         mpol_put(priv->task_mempolicy);
 109 }
 110 #else
 111 static void hold_task_mempolicy(struct proc_maps_private *priv)
 112 {
 113 }
 114 static void release_task_mempolicy(struct proc_maps_private *priv)
 115 {
 116 }
 117 #endif
 118
 119 static void vma_stop(struct proc_maps_private *priv)
 120 {
 121         struct mm_struct *mm = priv->mm;
 122
 123         release_task_mempolicy(priv);
 124         up_read(&mm->mmap_sem);
 125         mmput(mm);
 126 }
 127
 128 static struct vm_area_struct *
 129 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
 130 {
 131         if (vma == priv->tail_vma)
 132                 return NULL;
 133         return vma->vm_next ?: priv->tail_vma;
 134 }
 135
 136 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
 137 {
 138         if (m->count < m->size) /* vma is copied successfully */
 139                 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
 140 }
 141
 142 static void *m_start(struct seq_file *m, loff_t *ppos)
 143 {
 144         struct proc_maps_private *priv = m->private;
 145         unsigned long last_addr = m->version;
 146         struct mm_struct *mm;
 147         struct vm_area_struct *vma;
 148         unsigned int pos = *ppos;
 149
 150         /* See m_cache_vma(). Zero at the start or after lseek. */
 151         if (last_addr == -1UL)
 152                 return NULL;
 153
 154         priv->task = get_proc_task(priv->inode);
 155         if (!priv->task)
 156                 return ERR_PTR(-ESRCH);
 157
 158         mm = priv->mm;
 159         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
 160                 return NULL;
 161
 162         down_read(&mm->mmap_sem);
 163         hold_task_mempolicy(priv);
 164         priv->tail_vma = get_gate_vma(mm);
 165
 166         if (last_addr) {
 167                 vma = find_vma(mm, last_addr);
 168                 if (vma && (vma = m_next_vma(priv, vma)))
 169                         return vma;
 170         }
 171
 172         m->version = 0;
 173         if (pos < mm->map_count) {
 174                 for (vma = mm->mmap; pos; pos--) {
 175                         m->version = vma->vm_start;
 176                         vma = vma->vm_next;
 177                 }
 178                 return vma;
 179         }
 180
 181         /* we do not bother to update m->version in this case */
 182         if (pos == mm->map_count && priv->tail_vma)
 183                 return priv->tail_vma;
 184
 185         vma_stop(priv);
 186         return NULL;
 187 }
 188
 189 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 190 {
 191         struct proc_maps_private *priv = m->private;
 192         struct vm_area_struct *next;
 193
 194         (*pos)++;
 195         next = m_next_vma(priv, v);
 196         if (!next)
 197                 vma_stop(priv);
 198         return next;
 199 }
 200
 201 static void m_stop(struct seq_file *m, void *v)
 202 {
 203         struct proc_maps_private *priv = m->private;
 204
 205         if (!IS_ERR_OR_NULL(v))
 206                 vma_stop(priv);
 207         if (priv->task) {
 208                 put_task_struct(priv->task);
 209                 priv->task = NULL;
 210         }
 211 }
 212
 213 static int proc_maps_open(struct inode *inode, struct file *file,
 214                         const struct seq_operations *ops, int psize)
 215 {
 216         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 217
 218         if (!priv)
 219                 return -ENOMEM;
 220
 221         priv->inode = inode;
 222         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 223         if (IS_ERR(priv->mm)) {
 224                 int err = PTR_ERR(priv->mm);
 225
 226                 seq_release_private(inode, file);
 227                 return err;
 228         }
 229
 230         return 0;
 231 }
 232
 233 static int proc_map_release(struct inode *inode, struct file *file)
 234 {
 235         struct seq_file *seq = file->private_data;
 236         struct proc_maps_private *priv = seq->private;
 237
 238         if (priv->mm)
 239                 mmdrop(priv->mm);
 240
 241         return seq_release_private(inode, file);
 242 }
 243
 244 static int do_maps_open(struct inode *inode, struct file *file,
 245                         const struct seq_operations *ops)
 246 {
 247         return proc_maps_open(inode, file, ops,
 248                                 sizeof(struct proc_maps_private));
 249 }
 250
 251 static pid_t pid_of_stack(struct proc_maps_private *priv,
 252                                 struct vm_area_struct *vma, bool is_pid)
 253 {
 254         struct inode *inode = priv->inode;
 255         struct task_struct *task;
 256         pid_t ret = 0;
 257
 258         rcu_read_lock();
 259         task = pid_task(proc_pid(inode), PIDTYPE_PID);
 260         if (task) {
 261                 task = task_of_stack(task, vma, is_pid);
 262                 if (task)
 263                         ret = task_pid_nr_ns(task, inode->i_sb->s_fs_info);
 264         }
 265         rcu_read_unlock();
 266
 267         return ret;
 268 }
 269
 270 static void
 271 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 272 {
 273         struct mm_struct *mm = vma->vm_mm;
 274         struct file *file = vma->vm_file;
 275         struct proc_maps_private *priv = m->private;
 276         vm_flags_t flags = vma->vm_flags;
 277         unsigned long ino = 0;
 278         unsigned long long pgoff = 0;
 279         unsigned long start, end;
 280         dev_t dev = 0;
 281         const char *name = NULL;
 282
 283         if (file) {
 284                 struct inode *inode = file_inode(vma->vm_file);
 285                 dev = inode->i_sb->s_dev;
 286                 ino = inode->i_ino;
 287                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 288         }
 289
 290         /* We don't show the stack guard page in /proc/maps */
 291         start = vma->vm_start;
 292         if (stack_guard_page_start(vma, start))
 293                 start += PAGE_SIZE;
 294         end = vma->vm_end;
 295         if (stack_guard_page_end(vma, end))
 296                 end -= PAGE_SIZE;
 297
 298         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 299         seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
 300                         start,
 301                         end,
 302                         flags & VM_READ ? 'r' : '-',
 303                         flags & VM_WRITE ? 'w' : '-',
 304                         flags & VM_EXEC ? 'x' : '-',
 305                         flags & VM_MAYSHARE ? 's' : 'p',
 306                         pgoff,
 307                         MAJOR(dev), MINOR(dev), ino);
 308
 309         /*
 310          * Print the dentry name for named mappings, and a
 311          * special [heap] marker for the heap:
 312          */
 313         if (file) {
 314                 seq_pad(m, ' ');
 315                 seq_file_path(m, file, "\n");
 316                 goto done;
 317         }
 318
 319         if (vma->vm_ops && vma->vm_ops->name) {
 320                 name = vma->vm_ops->name(vma);
 321                 if (name)
 322                         goto done;
 323         }
 324
 325         name = arch_vma_name(vma);
 326         if (!name) {
 327                 pid_t tid;
 328
 329                 if (!mm) {
 330                         name = "[vdso]";
 331                         goto done;
 332                 }
 333
 334                 if (vma->vm_start <= mm->brk &&
 335                     vma->vm_end >= mm->start_brk) {
 336                         name = "[heap]";
 337                         goto done;
 338                 }
 339
 340                 tid = pid_of_stack(priv, vma, is_pid);
 341                 if (tid != 0) {
 342                         /*
 343                          * Thread stack in /proc/PID/task/TID/maps or
 344                          * the main process stack.
 345                          */
 346                         if (!is_pid || (vma->vm_start <= mm->start_stack &&
 347                             vma->vm_end >= mm->start_stack)) {
 348                                 name = "[stack]";
 349                         } else {
 350                                 /* Thread stack in /proc/PID/maps */
 351                                 seq_pad(m, ' ');
 352                                 seq_printf(m, "[stack:%d]", tid);
 353                         }
 354                 }
 355         }
 356
 357 done:
 358         if (name) {
 359                 seq_pad(m, ' ');
 360                 seq_puts(m, name);
 361         }
 362         seq_putc(m, '\n');
 363 }
 364
 365 static int show_map(struct seq_file *m, void *v, int is_pid)
 366 {
 367         show_map_vma(m, v, is_pid);
 368         m_cache_vma(m, v);
 369         return 0;
 370 }
 371
 372 static int show_pid_map(struct seq_file *m, void *v)
 373 {
 374         return show_map(m, v, 1);
 375 }
 376
 377 static int show_tid_map(struct seq_file *m, void *v)
 378 {
 379         return show_map(m, v, 0);
 380 }
 381
 382 static const struct seq_operations proc_pid_maps_op = {
 383         .start  = m_start,
 384         .next   = m_next,
 385         .stop   = m_stop,
 386         .show   = show_pid_map
 387 };
 388
 389 static const struct seq_operations proc_tid_maps_op = {
 390         .start  = m_start,
 391         .next   = m_next,
 392         .stop   = m_stop,
 393         .show   = show_tid_map
 394 };
 395
 396 static int pid_maps_open(struct inode *inode, struct file *file)
 397 {
 398         return do_maps_open(inode, file, &proc_pid_maps_op);
 399 }
 400
 401 static int tid_maps_open(struct inode *inode, struct file *file)
 402 {
 403         return do_maps_open(inode, file, &proc_tid_maps_op);
 404 }
 405
 406 const struct file_operations proc_pid_maps_operations = {
 407         .open           = pid_maps_open,
 408         .read           = seq_read,
 409         .llseek         = seq_lseek,
 410         .release        = proc_map_release,
 411 };
 412
 413 const struct file_operations proc_tid_maps_operations = {
 414         .open           = tid_maps_open,
 415         .read           = seq_read,
 416         .llseek         = seq_lseek,
 417         .release        = proc_map_release,
 418 };
 419
 420 /*
 421  * Proportional Set Size(PSS): my share of RSS.
 422  *
 423  * PSS of a process is the count of pages it has in memory, where each
 424  * page is divided by the number of processes sharing it.  So if a
 425  * process has 1000 pages all to itself, and 1000 shared with one other
 426  * process, its PSS will be 1500.
 427  *
 428  * To keep (accumulated) division errors low, we adopt a 64bit
 429  * fixed-point pss counter to minimize division errors. So (pss >>
 430  * PSS_SHIFT) would be the real byte count.
 431  *
 432  * A shift of 12 before division means (assuming 4K page size):
 433  *      - 1M 3-user-pages add up to 8KB errors;
 434  *      - supports mapcount up to 2^24, or 16M;
 435  *      - supports PSS up to 2^52 bytes, or 4PB.
 436  */
 437 #define PSS_SHIFT 12
 438
 439 #ifdef CONFIG_PROC_PAGE_MONITOR
 440 struct mem_size_stats {
 441         unsigned long resident;
 442         unsigned long shared_clean;
 443         unsigned long shared_dirty;
 444         unsigned long private_clean;
 445         unsigned long private_dirty;
 446         unsigned long referenced;
 447         unsigned long anonymous;
 448         unsigned long anonymous_thp;
 449         unsigned long swap;
 450         unsigned long shared_hugetlb;
 451         unsigned long private_hugetlb;
 452         u64 pss;
 453         u64 swap_pss;
 454 };
 455
 456 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 457                 bool compound, bool young, bool dirty)
 458 {
 459         int i, nr = compound ? HPAGE_PMD_NR : 1;
 460         unsigned long size = nr * PAGE_SIZE;
 461
 462         if (PageAnon(page))
 463                 mss->anonymous += size;
 464
 465         mss->resident += size;
 466         /* Accumulate the size in pages that have been accessed. */
 467         if (young || page_is_young(page) || PageReferenced(page))
 468                 mss->referenced += size;
 469
 470         /*
 471          * page_count(page) == 1 guarantees the page is mapped exactly once.
 472          * If any subpage of the compound page mapped with PTE it would elevate
 473          * page_count().
 474          */
 475         if (page_count(page) == 1) {
 476                 if (dirty || PageDirty(page))
 477                         mss->private_dirty += size;
 478                 else
 479                         mss->private_clean += size;
 480                 mss->pss += (u64)size << PSS_SHIFT;
 481                 return;
 482         }
 483
 484         for (i = 0; i < nr; i++, page++) {
 485                 int mapcount = page_mapcount(page);
 486
 487                 if (mapcount >= 2) {
 488                         if (dirty || PageDirty(page))
 489                                 mss->shared_dirty += PAGE_SIZE;
 490                         else
 491                                 mss->shared_clean += PAGE_SIZE;
 492                         mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
 493                 } else {
 494                         if (dirty || PageDirty(page))
 495                                 mss->private_dirty += PAGE_SIZE;
 496                         else
 497                                 mss->private_clean += PAGE_SIZE;
 498                         mss->pss += PAGE_SIZE << PSS_SHIFT;
 499                 }
 500         }
 501 }
 502
 503 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 504                 struct mm_walk *walk)
 505 {
 506         struct mem_size_stats *mss = walk->private;
 507         struct vm_area_struct *vma = walk->vma;
 508         struct page *page = NULL;
 509
 510         if (pte_present(*pte)) {
 511                 page = vm_normal_page(vma, addr, *pte);
 512         } else if (is_swap_pte(*pte)) {
 513                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 514
 515                 if (!non_swap_entry(swpent)) {
 516                         int mapcount;
 517
 518                         mss->swap += PAGE_SIZE;
 519                         mapcount = swp_swapcount(swpent);
 520                         if (mapcount >= 2) {
 521                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 522
 523                                 do_div(pss_delta, mapcount);
 524                                 mss->swap_pss += pss_delta;
 525                         } else {
 526                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 527                         }
 528                 } else if (is_migration_entry(swpent))
 529                         page = migration_entry_to_page(swpent);
 530         }
 531
 532         if (!page)
 533                 return;
 534
 535         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
 536 }
 537
 538 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 539 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 540                 struct mm_walk *walk)
 541 {
 542         struct mem_size_stats *mss = walk->private;
 543         struct vm_area_struct *vma = walk->vma;
 544         struct page *page;
 545
 546         /* FOLL_DUMP will return -EFAULT on huge zero page */
 547         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 548         if (IS_ERR_OR_NULL(page))
 549                 return;
 550         mss->anonymous_thp += HPAGE_PMD_SIZE;
 551         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
 552 }
 553 #else
 554 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 555                 struct mm_walk *walk)
 556 {
 557 }
 558 #endif
 559
 560 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 561                            struct mm_walk *walk)
 562 {
 563         struct vm_area_struct *vma = walk->vma;
 564         pte_t *pte;
 565         spinlock_t *ptl;
 566
 567         if (pmd_trans_huge_lock(pmd, vma, &ptl)) {
 568                 smaps_pmd_entry(pmd, addr, walk);
 569                 spin_unlock(ptl);
 570                 return 0;
 571         }
 572
 573         if (pmd_trans_unstable(pmd))
 574                 return 0;
 575         /*
 576          * The mmap_sem held all the way back in m_start() is what
 577          * keeps khugepaged out of here and from collapsing things
 578          * in here.
 579          */
 580         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 581         for (; addr != end; pte++, addr += PAGE_SIZE)
 582                 smaps_pte_entry(pte, addr, walk);
 583         pte_unmap_unlock(pte - 1, ptl);
 584         cond_resched();
 585         return 0;
 586 }
 587
 588 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 589 {
 590         /*
 591          * Don't forget to update Documentation/ on changes.
 592          */
 593         static const char mnemonics[BITS_PER_LONG][2] = {
 594                 /*
 595                  * In case if we meet a flag we don't know about.
 596                  */
 597                 [0 ... (BITS_PER_LONG-1)] = "??",
 598
 599                 [ilog2(VM_READ)]        = "rd",
 600                 [ilog2(VM_WRITE)]       = "wr",
 601                 [ilog2(VM_EXEC)]        = "ex",
 602                 [ilog2(VM_SHARED)]      = "sh",
 603                 [ilog2(VM_MAYREAD)]     = "mr",
 604                 [ilog2(VM_MAYWRITE)]    = "mw",
 605                 [ilog2(VM_MAYEXEC)]     = "me",
 606                 [ilog2(VM_MAYSHARE)]    = "ms",
 607                 [ilog2(VM_GROWSDOWN)]   = "gd",
 608                 [ilog2(VM_PFNMAP)]      = "pf",
 609                 [ilog2(VM_DENYWRITE)]   = "dw",
 610 #ifdef CONFIG_X86_INTEL_MPX
 611                 [ilog2(VM_MPX)]         = "mp",
 612 #endif
 613                 [ilog2(VM_LOCKED)]      = "lo",
 614                 [ilog2(VM_IO)]          = "io",
 615                 [ilog2(VM_SEQ_READ)]    = "sr",
 616                 [ilog2(VM_RAND_READ)]   = "rr",
 617                 [ilog2(VM_DONTCOPY)]    = "dc",
 618                 [ilog2(VM_DONTEXPAND)]  = "de",
 619                 [ilog2(VM_ACCOUNT)]     = "ac",
 620                 [ilog2(VM_NORESERVE)]   = "nr",
 621                 [ilog2(VM_HUGETLB)]     = "ht",
 622                 [ilog2(VM_ARCH_1)]      = "ar",
 623                 [ilog2(VM_DONTDUMP)]    = "dd",
 624 #ifdef CONFIG_MEM_SOFT_DIRTY
 625                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 626 #endif
 627                 [ilog2(VM_MIXEDMAP)]    = "mm",
 628                 [ilog2(VM_HUGEPAGE)]    = "hg",
 629                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 630                 [ilog2(VM_MERGEABLE)]   = "mg",
 631                 [ilog2(VM_UFFD_MISSING)]= "um",
 632                 [ilog2(VM_UFFD_WP)]     = "uw",
 633         };
 634         size_t i;
 635
 636         seq_puts(m, "VmFlags: ");
 637         for (i = 0; i < BITS_PER_LONG; i++) {
 638                 if (vma->vm_flags & (1UL << i)) {
 639                         seq_printf(m, "%c%c ",
 640                                    mnemonics[i][0], mnemonics[i][1]);
 641                 }
 642         }
 643         seq_putc(m, '\n');
 644 }
 645
 646 #ifdef CONFIG_HUGETLB_PAGE
 647 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 648                                  unsigned long addr, unsigned long end,
 649                                  struct mm_walk *walk)
 650 {
 651         struct mem_size_stats *mss = walk->private;
 652         struct vm_area_struct *vma = walk->vma;
 653         struct page *page = NULL;
 654
 655         if (pte_present(*pte)) {
 656                 page = vm_normal_page(vma, addr, *pte);
 657         } else if (is_swap_pte(*pte)) {
 658                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 659
 660                 if (is_migration_entry(swpent))
 661                         page = migration_entry_to_page(swpent);
 662         }
 663         if (page) {
 664                 int mapcount = page_mapcount(page);
 665
 666                 if (mapcount >= 2)
 667                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 668                 else
 669                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 670         }
 671         return 0;
 672 }
 673 #endif /* HUGETLB_PAGE */
 674
 675 static int show_smap(struct seq_file *m, void *v, int is_pid)
 676 {
 677         struct vm_area_struct *vma = v;
 678         struct mem_size_stats mss;
 679         struct mm_walk smaps_walk = {
 680                 .pmd_entry = smaps_pte_range,
 681 #ifdef CONFIG_HUGETLB_PAGE
 682                 .hugetlb_entry = smaps_hugetlb_range,
 683 #endif
 684                 .mm = vma->vm_mm,
 685                 .private = &mss,
 686         };
 687
 688         memset(&mss, 0, sizeof mss);
 689         /* mmap_sem is held in m_start */
 690         walk_page_vma(vma, &smaps_walk);
 691
 692         show_map_vma(m, vma, is_pid);
 693
 694         seq_printf(m,
 695                    "Size:           %8lu kB\n"
 696                    "Rss:            %8lu kB\n"
 697                    "Pss:            %8lu kB\n"
 698                    "Shared_Clean:   %8lu kB\n"
 699                    "Shared_Dirty:   %8lu kB\n"
 700                    "Private_Clean:  %8lu kB\n"
 701                    "Private_Dirty:  %8lu kB\n"
 702                    "Referenced:     %8lu kB\n"
 703                    "Anonymous:      %8lu kB\n"
 704                    "AnonHugePages:  %8lu kB\n"
 705                    "Shared_Hugetlb: %8lu kB\n"
 706                    "Private_Hugetlb: %8lu kB\n"
 707                    "Swap:           %8lu kB\n"
 708                    "SwapPss:        %8lu kB\n"
 709                    "KernelPageSize: %8lu kB\n"
 710                    "MMUPageSize:    %8lu kB\n"
 711                    "Locked:         %8lu kB\n",
 712                    (vma->vm_end - vma->vm_start) >> 10,
 713                    mss.resident >> 10,
 714                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 715                    mss.shared_clean  >> 10,
 716                    mss.shared_dirty  >> 10,
 717                    mss.private_clean >> 10,
 718                    mss.private_dirty >> 10,
 719                    mss.referenced >> 10,
 720                    mss.anonymous >> 10,
 721                    mss.anonymous_thp >> 10,
 722                    mss.shared_hugetlb >> 10,
 723                    mss.private_hugetlb >> 10,
 724                    mss.swap >> 10,
 725                    (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
 726                    vma_kernel_pagesize(vma) >> 10,
 727                    vma_mmu_pagesize(vma) >> 10,
 728                    (vma->vm_flags & VM_LOCKED) ?
 729                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 730
 731         show_smap_vma_flags(m, vma);
 732         m_cache_vma(m, vma);
 733         return 0;
 734 }
 735
 736 static int show_pid_smap(struct seq_file *m, void *v)
 737 {
 738         return show_smap(m, v, 1);
 739 }
 740
 741 static int show_tid_smap(struct seq_file *m, void *v)
 742 {
 743         return show_smap(m, v, 0);
 744 }
 745
 746 static const struct seq_operations proc_pid_smaps_op = {
 747         .start  = m_start,
 748         .next   = m_next,
 749         .stop   = m_stop,
 750         .show   = show_pid_smap
 751 };
 752
 753 static const struct seq_operations proc_tid_smaps_op = {
 754         .start  = m_start,
 755         .next   = m_next,
 756         .stop   = m_stop,
 757         .show   = show_tid_smap
 758 };
 759
 760 static int pid_smaps_open(struct inode *inode, struct file *file)
 761 {
 762         return do_maps_open(inode, file, &proc_pid_smaps_op);
 763 }
 764
 765 static int tid_smaps_open(struct inode *inode, struct file *file)
 766 {
 767         return do_maps_open(inode, file, &proc_tid_smaps_op);
 768 }
 769
 770 const struct file_operations proc_pid_smaps_operations = {
 771         .open           = pid_smaps_open,
 772         .read           = seq_read,
 773         .llseek         = seq_lseek,
 774         .release        = proc_map_release,
 775 };
 776
 777 const struct file_operations proc_tid_smaps_operations = {
 778         .open           = tid_smaps_open,
 779         .read           = seq_read,
 780         .llseek         = seq_lseek,
 781         .release        = proc_map_release,
 782 };
 783
 784 enum clear_refs_types {
 785         CLEAR_REFS_ALL = 1,
 786         CLEAR_REFS_ANON,
 787         CLEAR_REFS_MAPPED,
 788         CLEAR_REFS_SOFT_DIRTY,
 789         CLEAR_REFS_MM_HIWATER_RSS,
 790         CLEAR_REFS_LAST,
 791 };
 792
 793 struct clear_refs_private {
 794         enum clear_refs_types type;
 795 };
 796
 797 #ifdef CONFIG_MEM_SOFT_DIRTY
 798 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 799                 unsigned long addr, pte_t *pte)
 800 {
 801         /*
 802          * The soft-dirty tracker uses #PF-s to catch writes
 803          * to pages, so write-protect the pte as well. See the
 804          * Documentation/vm/soft-dirty.txt for full description
 805          * of how soft-dirty works.
 806          */
 807         pte_t ptent = *pte;
 808
 809         if (pte_present(ptent)) {
 810                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 811                 ptent = pte_wrprotect(ptent);
 812                 ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY);
 813                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 814         } else if (is_swap_pte(ptent)) {
 815                 ptent = pte_swp_clear_soft_dirty(ptent);
 816                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 817         }
 818 }
 819 #else
 820 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 821                 unsigned long addr, pte_t *pte)
 822 {
 823 }
 824 #endif
 825
 826 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 827 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 828                 unsigned long addr, pmd_t *pmdp)
 829 {
 830         pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
 831
 832         pmd = pmd_wrprotect(pmd);
 833         pmd = pmd_clear_flags(pmd, _PAGE_SOFT_DIRTY);
 834
 835         if (vma->vm_flags & VM_SOFTDIRTY)
 836                 vma->vm_flags &= ~VM_SOFTDIRTY;
 837
 838         set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 839 }
 840 #else
 841 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 842                 unsigned long addr, pmd_t *pmdp)
 843 {
 844 }
 845 #endif
 846
 847 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 848                                 unsigned long end, struct mm_walk *walk)
 849 {
 850         struct clear_refs_private *cp = walk->private;
 851         struct vm_area_struct *vma = walk->vma;
 852         pte_t *pte, ptent;
 853         spinlock_t *ptl;
 854         struct page *page;
 855
 856         if (pmd_trans_huge_lock(pmd, vma, &ptl)) {
 857                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 858                         clear_soft_dirty_pmd(vma, addr, pmd);
 859                         goto out;
 860                 }
 861
 862                 page = pmd_page(*pmd);
 863
 864                 /* Clear accessed and referenced bits. */
 865                 pmdp_test_and_clear_young(vma, addr, pmd);
 866                 test_and_clear_page_young(page);
 867                 ClearPageReferenced(page);
 868 out:
 869                 spin_unlock(ptl);
 870                 return 0;
 871         }
 872
 873         if (pmd_trans_unstable(pmd))
 874                 return 0;
 875
 876         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 877         for (; addr != end; pte++, addr += PAGE_SIZE) {
 878                 ptent = *pte;
 879
 880                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 881                         clear_soft_dirty(vma, addr, pte);
 882                         continue;
 883                 }
 884
 885                 if (!pte_present(ptent))
 886                         continue;
 887
 888                 page = vm_normal_page(vma, addr, ptent);
 889                 if (!page)
 890                         continue;
 891
 892                 /* Clear accessed and referenced bits. */
 893                 ptep_test_and_clear_young(vma, addr, pte);
 894                 test_and_clear_page_young(page);
 895                 ClearPageReferenced(page);
 896         }
 897         pte_unmap_unlock(pte - 1, ptl);
 898         cond_resched();
 899         return 0;
 900 }
 901
 902 static int clear_refs_test_walk(unsigned long start, unsigned long end,
 903                                 struct mm_walk *walk)
 904 {
 905         struct clear_refs_private *cp = walk->private;
 906         struct vm_area_struct *vma = walk->vma;
 907
 908         if (vma->vm_flags & VM_PFNMAP)
 909                 return 1;
 910
 911         /*
 912          * Writing 1 to /proc/pid/clear_refs affects all pages.
 913          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
 914          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
 915          * Writing 4 to /proc/pid/clear_refs affects all pages.
 916          */
 917         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
 918                 return 1;
 919         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
 920                 return 1;
 921         return 0;
 922 }
 923
 924 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 925                                 size_t count, loff_t *ppos)
 926 {
 927         struct task_struct *task;
 928         char buffer[PROC_NUMBUF];
 929         struct mm_struct *mm;
 930         struct vm_area_struct *vma;
 931         enum clear_refs_types type;
 932         int itype;
 933         int rv;
 934
 935         memset(buffer, 0, sizeof(buffer));
 936         if (count > sizeof(buffer) - 1)
 937                 count = sizeof(buffer) - 1;
 938         if (copy_from_user(buffer, buf, count))
 939                 return -EFAULT;
 940         rv = kstrtoint(strstrip(buffer), 10, &itype);
 941         if (rv < 0)
 942                 return rv;
 943         type = (enum clear_refs_types)itype;
 944         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
 945                 return -EINVAL;
 946
 947         task = get_proc_task(file_inode(file));
 948         if (!task)
 949                 return -ESRCH;
 950         mm = get_task_mm(task);
 951         if (mm) {
 952                 struct clear_refs_private cp = {
 953                         .type = type,
 954                 };
 955                 struct mm_walk clear_refs_walk = {
 956                         .pmd_entry = clear_refs_pte_range,
 957                         .test_walk = clear_refs_test_walk,
 958                         .mm = mm,
 959                         .private = &cp,
 960                 };
 961
 962                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
 963                         /*
 964                          * Writing 5 to /proc/pid/clear_refs resets the peak
 965                          * resident set size to this mm's current rss value.
 966                          */
 967                         down_write(&mm->mmap_sem);
 968                         reset_mm_hiwater_rss(mm);
 969                         up_write(&mm->mmap_sem);
 970                         goto out_mm;
 971                 }
 972
 973                 down_read(&mm->mmap_sem);
 974                 if (type == CLEAR_REFS_SOFT_DIRTY) {
 975                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
 976                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
 977                                         continue;
 978                                 up_read(&mm->mmap_sem);
 979                                 down_write(&mm->mmap_sem);
 980                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
 981                                         vma->vm_flags &= ~VM_SOFTDIRTY;
 982                                         vma_set_page_prot(vma);
 983                                 }
 984                                 downgrade_write(&mm->mmap_sem);
 985                                 break;
 986                         }
 987                         mmu_notifier_invalidate_range_start(mm, 0, -1);
 988                 }
 989                 walk_page_range(0, ~0UL, &clear_refs_walk);
 990                 if (type == CLEAR_REFS_SOFT_DIRTY)
 991                         mmu_notifier_invalidate_range_end(mm, 0, -1);
 992                 flush_tlb_mm(mm);
 993                 up_read(&mm->mmap_sem);
 994 out_mm:
 995                 mmput(mm);
 996         }
 997         put_task_struct(task);
 998
 999         return count;
1000 }
1001
1002 const struct file_operations proc_clear_refs_operations = {
1003         .write          = clear_refs_write,
1004         .llseek         = noop_llseek,
1005 };
1006
1007 typedef struct {
1008         u64 pme;
1009 } pagemap_entry_t;
1010
1011 struct pagemapread {
1012         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1013         pagemap_entry_t *buffer;
1014         bool show_pfn;
1015 };
1016
1017 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1018 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1019
1020 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1021 #define PM_PFRAME_BITS          55
1022 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1023 #define PM_SOFT_DIRTY           BIT_ULL(55)
1024 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1025 #define PM_FILE                 BIT_ULL(61)
1026 #define PM_SWAP                 BIT_ULL(62)
1027 #define PM_PRESENT              BIT_ULL(63)
1028
1029 #define PM_END_OF_BUFFER    1
1030
1031 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1032 {
1033         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1034 }
1035
1036 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1037                           struct pagemapread *pm)
1038 {
1039         pm->buffer[pm->pos++] = *pme;
1040         if (pm->pos >= pm->len)
1041                 return PM_END_OF_BUFFER;
1042         return 0;
1043 }
1044
1045 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1046                                 struct mm_walk *walk)
1047 {
1048         struct pagemapread *pm = walk->private;
1049         unsigned long addr = start;
1050         int err = 0;
1051
1052         while (addr < end) {
1053                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1054                 pagemap_entry_t pme = make_pme(0, 0);
1055                 /* End of address space hole, which we mark as non-present. */
1056                 unsigned long hole_end;
1057
1058                 if (vma)
1059                         hole_end = min(end, vma->vm_start);
1060                 else
1061                         hole_end = end;
1062
1063                 for (; addr < hole_end; addr += PAGE_SIZE) {
1064                         err = add_to_pagemap(addr, &pme, pm);
1065                         if (err)
1066                                 goto out;
1067                 }
1068
1069                 if (!vma)
1070                         break;
1071
1072                 /* Addresses in the VMA. */
1073                 if (vma->vm_flags & VM_SOFTDIRTY)
1074                         pme = make_pme(0, PM_SOFT_DIRTY);
1075                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1076                         err = add_to_pagemap(addr, &pme, pm);
1077                         if (err)
1078                                 goto out;
1079                 }
1080         }
1081 out:
1082         return err;
1083 }
1084
1085 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1086                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1087 {
1088         u64 frame = 0, flags = 0;
1089         struct page *page = NULL;
1090
1091         if (pte_present(pte)) {
1092                 if (pm->show_pfn)
1093                         frame = pte_pfn(pte);
1094                 flags |= PM_PRESENT;
1095                 page = vm_normal_page(vma, addr, pte);
1096                 if (pte_soft_dirty(pte))
1097                         flags |= PM_SOFT_DIRTY;
1098         } else if (is_swap_pte(pte)) {
1099                 swp_entry_t entry;
1100                 if (pte_swp_soft_dirty(pte))
1101                         flags |= PM_SOFT_DIRTY;
1102                 entry = pte_to_swp_entry(pte);
1103                 frame = swp_type(entry) |
1104                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1105                 flags |= PM_SWAP;
1106                 if (is_migration_entry(entry))
1107                         page = migration_entry_to_page(entry);
1108         }
1109
1110         if (page && !PageAnon(page))
1111                 flags |= PM_FILE;
1112         if (page && page_mapcount(page) == 1)
1113                 flags |= PM_MMAP_EXCLUSIVE;
1114         if (vma->vm_flags & VM_SOFTDIRTY)
1115                 flags |= PM_SOFT_DIRTY;
1116
1117         return make_pme(frame, flags);
1118 }
1119
1120 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1121                              struct mm_walk *walk)
1122 {
1123         struct vm_area_struct *vma = walk->vma;
1124         struct pagemapread *pm = walk->private;
1125         spinlock_t *ptl;
1126         pte_t *pte, *orig_pte;
1127         int err = 0;
1128
1129 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1130         if (pmd_trans_huge_lock(pmdp, vma, &ptl)) {
1131                 u64 flags = 0, frame = 0;
1132                 pmd_t pmd = *pmdp;
1133
1134                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1135                         flags |= PM_SOFT_DIRTY;
1136
1137                 /*
1138                  * Currently pmd for thp is always present because thp
1139                  * can not be swapped-out, migrated, or HWPOISONed
1140                  * (split in such cases instead.)
1141                  * This if-check is just to prepare for future implementation.
1142                  */
1143                 if (pmd_present(pmd)) {
1144                         struct page *page = pmd_page(pmd);
1145
1146                         if (page_mapcount(page) == 1)
1147                                 flags |= PM_MMAP_EXCLUSIVE;
1148
1149                         flags |= PM_PRESENT;
1150                         if (pm->show_pfn)
1151                                 frame = pmd_pfn(pmd) +
1152                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1153                 }
1154
1155                 for (; addr != end; addr += PAGE_SIZE) {
1156                         pagemap_entry_t pme = make_pme(frame, flags);
1157
1158                         err = add_to_pagemap(addr, &pme, pm);
1159                         if (err)
1160                                 break;
1161                         if (pm->show_pfn && (flags & PM_PRESENT))
1162                                 frame++;
1163                 }
1164                 spin_unlock(ptl);
1165                 return err;
1166         }
1167
1168         if (pmd_trans_unstable(pmdp))
1169                 return 0;
1170 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1171
1172         /*
1173          * We can assume that @vma always points to a valid one and @end never
1174          * goes beyond vma->vm_end.
1175          */
1176         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1177         for (; addr < end; pte++, addr += PAGE_SIZE) {
1178                 pagemap_entry_t pme;
1179
1180                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1181                 err = add_to_pagemap(addr, &pme, pm);
1182                 if (err)
1183                         break;
1184         }
1185         pte_unmap_unlock(orig_pte, ptl);
1186
1187         cond_resched();
1188
1189         return err;
1190 }
1191
1192 #ifdef CONFIG_HUGETLB_PAGE
1193 /* This function walks within one hugetlb entry in the single call */
1194 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1195                                  unsigned long addr, unsigned long end,
1196                                  struct mm_walk *walk)
1197 {
1198         struct pagemapread *pm = walk->private;
1199         struct vm_area_struct *vma = walk->vma;
1200         u64 flags = 0, frame = 0;
1201         int err = 0;
1202         pte_t pte;
1203
1204         if (vma->vm_flags & VM_SOFTDIRTY)
1205                 flags |= PM_SOFT_DIRTY;
1206
1207         pte = huge_ptep_get(ptep);
1208         if (pte_present(pte)) {
1209                 struct page *page = pte_page(pte);
1210
1211                 if (!PageAnon(page))
1212                         flags |= PM_FILE;
1213
1214                 if (page_mapcount(page) == 1)
1215                         flags |= PM_MMAP_EXCLUSIVE;
1216
1217                 flags |= PM_PRESENT;
1218                 if (pm->show_pfn)
1219                         frame = pte_pfn(pte) +
1220                                 ((addr & ~hmask) >> PAGE_SHIFT);
1221         }
1222
1223         for (; addr != end; addr += PAGE_SIZE) {
1224                 pagemap_entry_t pme = make_pme(frame, flags);
1225
1226                 err = add_to_pagemap(addr, &pme, pm);
1227                 if (err)
1228                         return err;
1229                 if (pm->show_pfn && (flags & PM_PRESENT))
1230                         frame++;
1231         }
1232
1233         cond_resched();
1234
1235         return err;
1236 }
1237 #endif /* HUGETLB_PAGE */
1238
1239 /*
1240  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1241  *
1242  * For each page in the address space, this file contains one 64-bit entry
1243  * consisting of the following:
1244  *
1245  * Bits 0-54  page frame number (PFN) if present
1246  * Bits 0-4   swap type if swapped
1247  * Bits 5-54  swap offset if swapped
1248  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1249  * Bit  56    page exclusively mapped
1250  * Bits 57-60 zero
1251  * Bit  61    page is file-page or shared-anon
1252  * Bit  62    page swapped
1253  * Bit  63    page present
1254  *
1255  * If the page is not present but in swap, then the PFN contains an
1256  * encoding of the swap file number and the page's offset into the
1257  * swap. Unmapped pages return a null PFN. This allows determining
1258  * precisely which pages are mapped (or in swap) and comparing mapped
1259  * pages between processes.
1260  *
1261  * Efficient users of this interface will use /proc/pid/maps to
1262  * determine which areas of memory are actually mapped and llseek to
1263  * skip over unmapped regions.
1264  */
1265 static ssize_t pagemap_read(struct file *file, char __user *buf,
1266                             size_t count, loff_t *ppos)
1267 {
1268         struct mm_struct *mm = file->private_data;
1269         struct pagemapread pm;
1270         struct mm_walk pagemap_walk = {};
1271         unsigned long src;
1272         unsigned long svpfn;
1273         unsigned long start_vaddr;
1274         unsigned long end_vaddr;
1275         int ret = 0, copied = 0;
1276
1277         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1278                 goto out;
1279
1280         ret = -EINVAL;
1281         /* file position must be aligned */
1282         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1283                 goto out_mm;
1284
1285         ret = 0;
1286         if (!count)
1287                 goto out_mm;
1288
1289         /* do not disclose physical addresses: attack vector */
1290         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1291
1292         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1293         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1294         ret = -ENOMEM;
1295         if (!pm.buffer)
1296                 goto out_mm;
1297
1298         pagemap_walk.pmd_entry = pagemap_pmd_range;
1299         pagemap_walk.pte_hole = pagemap_pte_hole;
1300 #ifdef CONFIG_HUGETLB_PAGE
1301         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1302 #endif
1303         pagemap_walk.mm = mm;
1304         pagemap_walk.private = &pm;
1305
1306         src = *ppos;
1307         svpfn = src / PM_ENTRY_BYTES;
1308         start_vaddr = svpfn << PAGE_SHIFT;
1309         end_vaddr = mm->task_size;
1310
1311         /* watch out for wraparound */
1312         if (svpfn > mm->task_size >> PAGE_SHIFT)
1313                 start_vaddr = end_vaddr;
1314
1315         /*
1316          * The odds are that this will stop walking way
1317          * before end_vaddr, because the length of the
1318          * user buffer is tracked in "pm", and the walk
1319          * will stop when we hit the end of the buffer.
1320          */
1321         ret = 0;
1322         while (count && (start_vaddr < end_vaddr)) {
1323                 int len;
1324                 unsigned long end;
1325
1326                 pm.pos = 0;
1327                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1328                 /* overflow ? */
1329                 if (end < start_vaddr || end > end_vaddr)
1330                         end = end_vaddr;
1331                 down_read(&mm->mmap_sem);
1332                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1333                 up_read(&mm->mmap_sem);
1334                 start_vaddr = end;
1335
1336                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1337                 if (copy_to_user(buf, pm.buffer, len)) {
1338                         ret = -EFAULT;
1339                         goto out_free;
1340                 }
1341                 copied += len;
1342                 buf += len;
1343                 count -= len;
1344         }
1345         *ppos += copied;
1346         if (!ret || ret == PM_END_OF_BUFFER)
1347                 ret = copied;
1348
1349 out_free:
1350         kfree(pm.buffer);
1351 out_mm:
1352         mmput(mm);
1353 out:
1354         return ret;
1355 }
1356
1357 static int pagemap_open(struct inode *inode, struct file *file)
1358 {
1359         struct mm_struct *mm;
1360
1361         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1362         if (IS_ERR(mm))
1363                 return PTR_ERR(mm);
1364         file->private_data = mm;
1365         return 0;
1366 }
1367
1368 static int pagemap_release(struct inode *inode, struct file *file)
1369 {
1370         struct mm_struct *mm = file->private_data;
1371
1372         if (mm)
1373                 mmdrop(mm);
1374         return 0;
1375 }
1376
1377 const struct file_operations proc_pagemap_operations = {
1378         .llseek         = mem_lseek, /* borrow this */
1379         .read           = pagemap_read,
1380         .open           = pagemap_open,
1381         .release        = pagemap_release,
1382 };
1383 #endif /* CONFIG_PROC_PAGE_MONITOR */
1384
1385 #ifdef CONFIG_NUMA
1386
1387 struct numa_maps {
1388         unsigned long pages;
1389         unsigned long anon;
1390         unsigned long active;
1391         unsigned long writeback;
1392         unsigned long mapcount_max;
1393         unsigned long dirty;
1394         unsigned long swapcache;
1395         unsigned long node[MAX_NUMNODES];
1396 };
1397
1398 struct numa_maps_private {
1399         struct proc_maps_private proc_maps;
1400         struct numa_maps md;
1401 };
1402
1403 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1404                         unsigned long nr_pages)
1405 {
1406         int count = page_mapcount(page);
1407
1408         md->pages += nr_pages;
1409         if (pte_dirty || PageDirty(page))
1410                 md->dirty += nr_pages;
1411
1412         if (PageSwapCache(page))
1413                 md->swapcache += nr_pages;
1414
1415         if (PageActive(page) || PageUnevictable(page))
1416                 md->active += nr_pages;
1417
1418         if (PageWriteback(page))
1419                 md->writeback += nr_pages;
1420
1421         if (PageAnon(page))
1422                 md->anon += nr_pages;
1423
1424         if (count > md->mapcount_max)
1425                 md->mapcount_max = count;
1426
1427         md->node[page_to_nid(page)] += nr_pages;
1428 }
1429
1430 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1431                 unsigned long addr)
1432 {
1433         struct page *page;
1434         int nid;
1435
1436         if (!pte_present(pte))
1437                 return NULL;
1438
1439         page = vm_normal_page(vma, addr, pte);
1440         if (!page)
1441                 return NULL;
1442
1443         if (PageReserved(page))
1444                 return NULL;
1445
1446         nid = page_to_nid(page);
1447         if (!node_isset(nid, node_states[N_MEMORY]))
1448                 return NULL;
1449
1450         return page;
1451 }
1452
1453 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1454                 unsigned long end, struct mm_walk *walk)
1455 {
1456         struct numa_maps *md = walk->private;
1457         struct vm_area_struct *vma = walk->vma;
1458         spinlock_t *ptl;
1459         pte_t *orig_pte;
1460         pte_t *pte;
1461
1462         if (pmd_trans_huge_lock(pmd, vma, &ptl)) {
1463                 pte_t huge_pte = *(pte_t *)pmd;
1464                 struct page *page;
1465
1466                 page = can_gather_numa_stats(huge_pte, vma, addr);
1467                 if (page)
1468                         gather_stats(page, md, pte_dirty(huge_pte),
1469                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1470                 spin_unlock(ptl);
1471                 return 0;
1472         }
1473
1474         if (pmd_trans_unstable(pmd))
1475                 return 0;
1476         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1477         do {
1478                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1479                 if (!page)
1480                         continue;
1481                 gather_stats(page, md, pte_dirty(*pte), 1);
1482
1483         } while (pte++, addr += PAGE_SIZE, addr != end);
1484         pte_unmap_unlock(orig_pte, ptl);
1485         return 0;
1486 }
1487 #ifdef CONFIG_HUGETLB_PAGE
1488 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1489                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1490 {
1491         struct numa_maps *md;
1492         struct page *page;
1493
1494         if (!pte_present(*pte))
1495                 return 0;
1496
1497         page = pte_page(*pte);
1498         if (!page)
1499                 return 0;
1500
1501         md = walk->private;
1502         gather_stats(page, md, pte_dirty(*pte), 1);
1503         return 0;
1504 }
1505
1506 #else
1507 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1508                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1509 {
1510         return 0;
1511 }
1512 #endif
1513
1514 /*
1515  * Display pages allocated per node and memory policy via /proc.
1516  */
1517 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1518 {
1519         struct numa_maps_private *numa_priv = m->private;
1520         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1521         struct vm_area_struct *vma = v;
1522         struct numa_maps *md = &numa_priv->md;
1523         struct file *file = vma->vm_file;
1524         struct mm_struct *mm = vma->vm_mm;
1525         struct mm_walk walk = {
1526                 .hugetlb_entry = gather_hugetlb_stats,
1527                 .pmd_entry = gather_pte_stats,
1528                 .private = md,
1529                 .mm = mm,
1530         };
1531         struct mempolicy *pol;
1532         char buffer[64];
1533         int nid;
1534
1535         if (!mm)
1536                 return 0;
1537
1538         /* Ensure we start with an empty set of numa_maps statistics. */
1539         memset(md, 0, sizeof(*md));
1540
1541         pol = __get_vma_policy(vma, vma->vm_start);
1542         if (pol) {
1543                 mpol_to_str(buffer, sizeof(buffer), pol);
1544                 mpol_cond_put(pol);
1545         } else {
1546                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1547         }
1548
1549         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1550
1551         if (file) {
1552                 seq_puts(m, " file=");
1553                 seq_file_path(m, file, "\n\t= ");
1554         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1555                 seq_puts(m, " heap");
1556         } else {
1557                 pid_t tid = pid_of_stack(proc_priv, vma, is_pid);
1558                 if (tid != 0) {
1559                         /*
1560                          * Thread stack in /proc/PID/task/TID/maps or
1561                          * the main process stack.
1562                          */
1563                         if (!is_pid || (vma->vm_start <= mm->start_stack &&
1564                             vma->vm_end >= mm->start_stack))
1565                                 seq_puts(m, " stack");
1566                         else
1567                                 seq_printf(m, " stack:%d", tid);
1568                 }
1569         }
1570
1571         if (is_vm_hugetlb_page(vma))
1572                 seq_puts(m, " huge");
1573
1574         /* mmap_sem is held by m_start */
1575         walk_page_vma(vma, &walk);
1576
1577         if (!md->pages)
1578                 goto out;
1579
1580         if (md->anon)
1581                 seq_printf(m, " anon=%lu", md->anon);
1582
1583         if (md->dirty)
1584                 seq_printf(m, " dirty=%lu", md->dirty);
1585
1586         if (md->pages != md->anon && md->pages != md->dirty)
1587                 seq_printf(m, " mapped=%lu", md->pages);
1588
1589         if (md->mapcount_max > 1)
1590                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1591
1592         if (md->swapcache)
1593                 seq_printf(m, " swapcache=%lu", md->swapcache);
1594
1595         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1596                 seq_printf(m, " active=%lu", md->active);
1597
1598         if (md->writeback)
1599                 seq_printf(m, " writeback=%lu", md->writeback);
1600
1601         for_each_node_state(nid, N_MEMORY)
1602                 if (md->node[nid])
1603                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1604
1605         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1606 out:
1607         seq_putc(m, '\n');
1608         m_cache_vma(m, vma);
1609         return 0;
1610 }
1611
1612 static int show_pid_numa_map(struct seq_file *m, void *v)
1613 {
1614         return show_numa_map(m, v, 1);
1615 }
1616
1617 static int show_tid_numa_map(struct seq_file *m, void *v)
1618 {
1619         return show_numa_map(m, v, 0);
1620 }
1621
1622 static const struct seq_operations proc_pid_numa_maps_op = {
1623         .start  = m_start,
1624         .next   = m_next,
1625         .stop   = m_stop,
1626         .show   = show_pid_numa_map,
1627 };
1628
1629 static const struct seq_operations proc_tid_numa_maps_op = {
1630         .start  = m_start,
1631         .next   = m_next,
1632         .stop   = m_stop,
1633         .show   = show_tid_numa_map,
1634 };
1635
1636 static int numa_maps_open(struct inode *inode, struct file *file,
1637                           const struct seq_operations *ops)
1638 {
1639         return proc_maps_open(inode, file, ops,
1640                                 sizeof(struct numa_maps_private));
1641 }
1642
1643 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1644 {
1645         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1646 }
1647
1648 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1649 {
1650         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1651 }
1652
1653 const struct file_operations proc_pid_numa_maps_operations = {
1654         .open           = pid_numa_maps_open,
1655         .read           = seq_read,
1656         .llseek         = seq_lseek,
1657         .release        = proc_map_release,
1658 };
1659
1660 const struct file_operations proc_tid_numa_maps_operations = {
1661         .open           = tid_numa_maps_open,
1662         .read           = seq_read,
1663         .llseek         = seq_lseek,
1664         .release        = proc_map_release,
1665 };
1666 #endif /* CONFIG_NUMA */