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