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