kernel/fs/proc/task_mmu.c

   1 #include <linux/mm.h>
   2 #include <linux/vmacache.h>
   3 #include <linux/hugetlb.h>
   4 #include <linux/huge_mm.h>
   5 #include <linux/mount.h>
   6 #include <linux/seq_file.h>
   7 #include <linux/highmem.h>
   8 #include <linux/ptrace.h>
   9 #include <linux/slab.h>
  10 #include <linux/pagemap.h>
  11 #include <linux/mempolicy.h>
  12 #include <linux/rmap.h>
  13 #include <linux/swap.h>
  14 #include <linux/swapops.h>
  15 #include <linux/mmu_notifier.h>
  16 #include <linux/page_idle.h>
  17
  18 #include <asm/elf.h>
  19 #include <asm/uaccess.h>
  20 #include <asm/tlbflush.h>
  21 #include "internal.h"
  22
  23 void task_mem(struct seq_file *m, struct mm_struct *mm)
  24 {
  25         unsigned long data, text, lib, swap, ptes, pmds;
  26         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  27
  28         /*
  29          * Note: to minimize their overhead, mm maintains hiwater_vm and
  30          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  31          * collector of these hiwater stats must therefore get total_vm
  32          * and rss too, which will usually be the higher.  Barriers? not
  33          * worth the effort, such snapshots can always be inconsistent.
  34          */
  35         hiwater_vm = total_vm = mm->total_vm;
  36         if (hiwater_vm < mm->hiwater_vm)
  37                 hiwater_vm = mm->hiwater_vm;
  38         hiwater_rss = total_rss = get_mm_rss(mm);
  39         if (hiwater_rss < mm->hiwater_rss)
  40                 hiwater_rss = mm->hiwater_rss;
  41
  42         data = mm->total_vm - mm->shared_vm - mm->stack_vm;
  43         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  44         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  45         swap = get_mm_counter(mm, MM_SWAPENTS);
  46         ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
  47         pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
  48         seq_printf(m,
  49                 "VmPeak:\t%8lu kB\n"
  50                 "VmSize:\t%8lu kB\n"
  51                 "VmLck:\t%8lu kB\n"
  52                 "VmPin:\t%8lu kB\n"
  53                 "VmHWM:\t%8lu kB\n"
  54                 "VmRSS:\t%8lu kB\n"
  55                 "VmData:\t%8lu kB\n"
  56                 "VmStk:\t%8lu kB\n"
  57                 "VmExe:\t%8lu kB\n"
  58                 "VmLib:\t%8lu kB\n"
  59                 "VmPTE:\t%8lu kB\n"
  60                 "VmPMD:\t%8lu kB\n"
  61                 "VmSwap:\t%8lu kB\n",
  62                 hiwater_vm << (PAGE_SHIFT-10),
  63                 total_vm << (PAGE_SHIFT-10),
  64                 mm->locked_vm << (PAGE_SHIFT-10),
  65                 mm->pinned_vm << (PAGE_SHIFT-10),
  66                 hiwater_rss << (PAGE_SHIFT-10),
  67                 total_rss << (PAGE_SHIFT-10),
  68                 data << (PAGE_SHIFT-10),
  69                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  70                 ptes >> 10,
  71                 pmds >> 10,
  72                 swap << (PAGE_SHIFT-10));
  73         hugetlb_report_usage(m, mm);
  74 }
  75
  76 unsigned long task_vsize(struct mm_struct *mm)
  77 {
  78         return PAGE_SIZE * mm->total_vm;
  79 }
  80
  81 unsigned long task_statm(struct mm_struct *mm,
  82                          unsigned long *shared, unsigned long *text,
  83                          unsigned long *data, unsigned long *resident)
  84 {
  85         *shared = get_mm_counter(mm, MM_FILEPAGES);
  86         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  87                                                                 >> PAGE_SHIFT;
  88         *data = mm->total_vm - mm->shared_vm;
  89         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
  90         return mm->total_vm;
  91 }
  92
  93 #ifdef CONFIG_NUMA
  94 /*
  95  * Save get_task_policy() for show_numa_map().
  96  */
  97 static void hold_task_mempolicy(struct proc_maps_private *priv)
  98 {
  99         struct task_struct *task = priv->task;
 100
 101         task_lock(task);
 102         priv->task_mempolicy = get_task_policy(task);
 103         mpol_get(priv->task_mempolicy);
 104         task_unlock(task);
 105 }
 106 static void release_task_mempolicy(struct proc_maps_private *priv)
 107 {
 108         mpol_put(priv->task_mempolicy);
 109 }
 110 #else
 111 static void hold_task_mempolicy(struct proc_maps_private *priv)
 112 {
 113 }
 114 static void release_task_mempolicy(struct proc_maps_private *priv)
 115 {
 116 }
 117 #endif
 118
 119 static void vma_stop(struct proc_maps_private *priv)
 120 {
 121         struct mm_struct *mm = priv->mm;
 122
 123         release_task_mempolicy(priv);
 124         up_read(&mm->mmap_sem);
 125         mmput(mm);
 126 }
 127
 128 static struct vm_area_struct *
 129 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
 130 {
 131         if (vma == priv->tail_vma)
 132                 return NULL;
 133         return vma->vm_next ?: priv->tail_vma;
 134 }
 135
 136 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
 137 {
 138         if (m->count < m->size) /* vma is copied successfully */
 139                 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
 140 }
 141
 142 static void *m_start(struct seq_file *m, loff_t *ppos)
 143 {
 144         struct proc_maps_private *priv = m->private;
 145         unsigned long last_addr = m->version;
 146         struct mm_struct *mm;
 147         struct vm_area_struct *vma;
 148         unsigned int pos = *ppos;
 149
 150         /* See m_cache_vma(). Zero at the start or after lseek. */
 151         if (last_addr == -1UL)
 152                 return NULL;
 153
 154         priv->task = get_proc_task(priv->inode);
 155         if (!priv->task)
 156                 return ERR_PTR(-ESRCH);
 157
 158         mm = priv->mm;
 159         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
 160                 return NULL;
 161
 162         down_read(&mm->mmap_sem);
 163         hold_task_mempolicy(priv);
 164         priv->tail_vma = get_gate_vma(mm);
 165
 166         if (last_addr) {
 167                 vma = find_vma(mm, last_addr);
 168                 if (vma && (vma = m_next_vma(priv, vma)))
 169                         return vma;
 170         }
 171
 172         m->version = 0;
 173         if (pos < mm->map_count) {
 174                 for (vma = mm->mmap; pos; pos--) {
 175                         m->version = vma->vm_start;
 176                         vma = vma->vm_next;
 177                 }
 178                 return vma;
 179         }
 180
 181         /* we do not bother to update m->version in this case */
 182         if (pos == mm->map_count && priv->tail_vma)
 183                 return priv->tail_vma;
 184
 185         vma_stop(priv);
 186         return NULL;
 187 }
 188
 189 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 190 {
 191         struct proc_maps_private *priv = m->private;
 192         struct vm_area_struct *next;
 193
 194         (*pos)++;
 195         next = m_next_vma(priv, v);
 196         if (!next)
 197                 vma_stop(priv);
 198         return next;
 199 }
 200
 201 static void m_stop(struct seq_file *m, void *v)
 202 {
 203         struct proc_maps_private *priv = m->private;
 204
 205         if (!IS_ERR_OR_NULL(v))
 206                 vma_stop(priv);
 207         if (priv->task) {
 208                 put_task_struct(priv->task);
 209                 priv->task = NULL;
 210         }
 211 }
 212
 213 static int proc_maps_open(struct inode *inode, struct file *file,
 214                         const struct seq_operations *ops, int psize)
 215 {
 216         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 217
 218         if (!priv)
 219                 return -ENOMEM;
 220
 221         priv->inode = inode;
 222         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 223         if (IS_ERR(priv->mm)) {
 224                 int err = PTR_ERR(priv->mm);
 225
 226                 seq_release_private(inode, file);
 227                 return err;
 228         }
 229
 230         return 0;
 231 }
 232
 233 static int proc_map_release(struct inode *inode, struct file *file)
 234 {
 235         struct seq_file *seq = file->private_data;
 236         struct proc_maps_private *priv = seq->private;
 237
 238         if (priv->mm)
 239                 mmdrop(priv->mm);
 240
 241         return seq_release_private(inode, file);
 242 }
 243
 244 static int do_maps_open(struct inode *inode, struct file *file,
 245                         const struct seq_operations *ops)
 246 {
 247         return proc_maps_open(inode, file, ops,
 248                                 sizeof(struct proc_maps_private));
 249 }
 250
 251 /*
 252  * Indicate if the VMA is a stack for the given task; for
 253  * /proc/PID/maps that is the stack of the main task.
 254  */
 255 static int is_stack(struct proc_maps_private *priv,
 256                     struct vm_area_struct *vma, int is_pid)
 257 {
 258         int stack = 0;
 259
 260         if (is_pid) {
 261                 stack = vma->vm_start <= vma->vm_mm->start_stack &&
 262                         vma->vm_end >= vma->vm_mm->start_stack;
 263         } else {
 264                 struct inode *inode = priv->inode;
 265                 struct task_struct *task;
 266
 267                 rcu_read_lock();
 268                 task = pid_task(proc_pid(inode), PIDTYPE_PID);
 269                 if (task)
 270                         stack = vma_is_stack_for_task(vma, task);
 271                 rcu_read_unlock();
 272         }
 273         return stack;
 274 }
 275
 276 static void
 277 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 278 {
 279         struct mm_struct *mm = vma->vm_mm;
 280         struct file *file = vma->vm_file;
 281         struct proc_maps_private *priv = m->private;
 282         vm_flags_t flags = vma->vm_flags;
 283         unsigned long ino = 0;
 284         unsigned long long pgoff = 0;
 285         unsigned long start, end;
 286         dev_t dev = 0;
 287         const char *name = NULL;
 288
 289         if (file) {
 290                 struct inode *inode = file_inode(vma->vm_file);
 291                 dev = inode->i_sb->s_dev;
 292                 ino = inode->i_ino;
 293                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 294         }
 295
 296         /* We don't show the stack guard page in /proc/maps */
 297         start = vma->vm_start;
 298         if (stack_guard_page_start(vma, start))
 299                 start += PAGE_SIZE;
 300         end = vma->vm_end;
 301         if (stack_guard_page_end(vma, end))
 302                 end -= PAGE_SIZE;
 303
 304         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 305         seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
 306                         start,
 307                         end,
 308                         flags & VM_READ ? 'r' : '-',
 309                         flags & VM_WRITE ? 'w' : '-',
 310                         flags & VM_EXEC ? 'x' : '-',
 311                         flags & VM_MAYSHARE ? 's' : 'p',
 312                         pgoff,
 313                         MAJOR(dev), MINOR(dev), ino);
 314
 315         /*
 316          * Print the dentry name for named mappings, and a
 317          * special [heap] marker for the heap:
 318          */
 319         if (file) {
 320                 seq_pad(m, ' ');
 321                 seq_file_path(m, file, "\n");
 322                 goto done;
 323         }
 324
 325         if (vma->vm_ops && vma->vm_ops->name) {
 326                 name = vma->vm_ops->name(vma);
 327                 if (name)
 328                         goto done;
 329         }
 330
 331         name = arch_vma_name(vma);
 332         if (!name) {
 333                 if (!mm) {
 334                         name = "[vdso]";
 335                         goto done;
 336                 }
 337
 338                 if (vma->vm_start <= mm->brk &&
 339                     vma->vm_end >= mm->start_brk) {
 340                         name = "[heap]";
 341                         goto done;
 342                 }
 343
 344                 if (is_stack(priv, vma, is_pid))
 345                         name = "[stack]";
 346         }
 347
 348 done:
 349         if (name) {
 350                 seq_pad(m, ' ');
 351                 seq_puts(m, name);
 352         }
 353         seq_putc(m, '\n');
 354 }
 355
 356 static int show_map(struct seq_file *m, void *v, int is_pid)
 357 {
 358         show_map_vma(m, v, is_pid);
 359         m_cache_vma(m, v);
 360         return 0;
 361 }
 362
 363 static int show_pid_map(struct seq_file *m, void *v)
 364 {
 365         return show_map(m, v, 1);
 366 }
 367
 368 static int show_tid_map(struct seq_file *m, void *v)
 369 {
 370         return show_map(m, v, 0);
 371 }
 372
 373 static const struct seq_operations proc_pid_maps_op = {
 374         .start  = m_start,
 375         .next   = m_next,
 376         .stop   = m_stop,
 377         .show   = show_pid_map
 378 };
 379
 380 static const struct seq_operations proc_tid_maps_op = {
 381         .start  = m_start,
 382         .next   = m_next,
 383         .stop   = m_stop,
 384         .show   = show_tid_map
 385 };
 386
 387 static int pid_maps_open(struct inode *inode, struct file *file)
 388 {
 389         return do_maps_open(inode, file, &proc_pid_maps_op);
 390 }
 391
 392 static int tid_maps_open(struct inode *inode, struct file *file)
 393 {
 394         return do_maps_open(inode, file, &proc_tid_maps_op);
 395 }
 396
 397 const struct file_operations proc_pid_maps_operations = {
 398         .open           = pid_maps_open,
 399         .read           = seq_read,
 400         .llseek         = seq_lseek,
 401         .release        = proc_map_release,
 402 };
 403
 404 const struct file_operations proc_tid_maps_operations = {
 405         .open           = tid_maps_open,
 406         .read           = seq_read,
 407         .llseek         = seq_lseek,
 408         .release        = proc_map_release,
 409 };
 410
 411 /*
 412  * Proportional Set Size(PSS): my share of RSS.
 413  *
 414  * PSS of a process is the count of pages it has in memory, where each
 415  * page is divided by the number of processes sharing it.  So if a
 416  * process has 1000 pages all to itself, and 1000 shared with one other
 417  * process, its PSS will be 1500.
 418  *
 419  * To keep (accumulated) division errors low, we adopt a 64bit
 420  * fixed-point pss counter to minimize division errors. So (pss >>
 421  * PSS_SHIFT) would be the real byte count.
 422  *
 423  * A shift of 12 before division means (assuming 4K page size):
 424  *      - 1M 3-user-pages add up to 8KB errors;
 425  *      - supports mapcount up to 2^24, or 16M;
 426  *      - supports PSS up to 2^52 bytes, or 4PB.
 427  */
 428 #define PSS_SHIFT 12
 429
 430 #ifdef CONFIG_PROC_PAGE_MONITOR
 431 struct mem_size_stats {
 432         unsigned long resident;
 433         unsigned long shared_clean;
 434         unsigned long shared_dirty;
 435         unsigned long private_clean;
 436         unsigned long private_dirty;
 437         unsigned long referenced;
 438         unsigned long anonymous;
 439         unsigned long anonymous_thp;
 440         unsigned long swap;
 441         unsigned long shared_hugetlb;
 442         unsigned long private_hugetlb;
 443         u64 pss;
 444         u64 swap_pss;
 445 };
 446
 447 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 448                 unsigned long size, bool young, bool dirty)
 449 {
 450         int mapcount;
 451
 452         if (PageAnon(page))
 453                 mss->anonymous += size;
 454
 455         mss->resident += size;
 456         /* Accumulate the size in pages that have been accessed. */
 457         if (young || page_is_young(page) || PageReferenced(page))
 458                 mss->referenced += size;
 459         mapcount = page_mapcount(page);
 460         if (mapcount >= 2) {
 461                 u64 pss_delta;
 462
 463                 if (dirty || PageDirty(page))
 464                         mss->shared_dirty += size;
 465                 else
 466                         mss->shared_clean += size;
 467                 pss_delta = (u64)size << PSS_SHIFT;
 468                 do_div(pss_delta, mapcount);
 469                 mss->pss += pss_delta;
 470         } else {
 471                 if (dirty || PageDirty(page))
 472                         mss->private_dirty += size;
 473                 else
 474                         mss->private_clean += size;
 475                 mss->pss += (u64)size << PSS_SHIFT;
 476         }
 477 }
 478
 479 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 480                 struct mm_walk *walk)
 481 {
 482         struct mem_size_stats *mss = walk->private;
 483         struct vm_area_struct *vma = walk->vma;
 484         struct page *page = NULL;
 485
 486         if (pte_present(*pte)) {
 487                 page = vm_normal_page(vma, addr, *pte);
 488         } else if (is_swap_pte(*pte)) {
 489                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 490
 491                 if (!non_swap_entry(swpent)) {
 492                         int mapcount;
 493
 494                         mss->swap += PAGE_SIZE;
 495                         mapcount = swp_swapcount(swpent);
 496                         if (mapcount >= 2) {
 497                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 498
 499                                 do_div(pss_delta, mapcount);
 500                                 mss->swap_pss += pss_delta;
 501                         } else {
 502                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 503                         }
 504                 } else if (is_migration_entry(swpent))
 505                         page = migration_entry_to_page(swpent);
 506         }
 507
 508         if (!page)
 509                 return;
 510         smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte));
 511 }
 512
 513 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 514 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 515                 struct mm_walk *walk)
 516 {
 517         struct mem_size_stats *mss = walk->private;
 518         struct vm_area_struct *vma = walk->vma;
 519         struct page *page;
 520
 521         /* FOLL_DUMP will return -EFAULT on huge zero page */
 522         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 523         if (IS_ERR_OR_NULL(page))
 524                 return;
 525         mss->anonymous_thp += HPAGE_PMD_SIZE;
 526         smaps_account(mss, page, HPAGE_PMD_SIZE,
 527                         pmd_young(*pmd), pmd_dirty(*pmd));
 528 }
 529 #else
 530 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 531                 struct mm_walk *walk)
 532 {
 533 }
 534 #endif
 535
 536 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 537                            struct mm_walk *walk)
 538 {
 539         struct vm_area_struct *vma = walk->vma;
 540         pte_t *pte;
 541         spinlock_t *ptl;
 542
 543         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
 544                 smaps_pmd_entry(pmd, addr, walk);
 545                 spin_unlock(ptl);
 546                 return 0;
 547         }
 548
 549         if (pmd_trans_unstable(pmd))
 550                 return 0;
 551         /*
 552          * The mmap_sem held all the way back in m_start() is what
 553          * keeps khugepaged out of here and from collapsing things
 554          * in here.
 555          */
 556         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 557         for (; addr != end; pte++, addr += PAGE_SIZE)
 558                 smaps_pte_entry(pte, addr, walk);
 559         pte_unmap_unlock(pte - 1, ptl);
 560         cond_resched();
 561         return 0;
 562 }
 563
 564 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 565 {
 566         /*
 567          * Don't forget to update Documentation/ on changes.
 568          */
 569         static const char mnemonics[BITS_PER_LONG][2] = {
 570                 /*
 571                  * In case if we meet a flag we don't know about.
 572                  */
 573                 [0 ... (BITS_PER_LONG-1)] = "??",
 574
 575                 [ilog2(VM_READ)]        = "rd",
 576                 [ilog2(VM_WRITE)]       = "wr",
 577                 [ilog2(VM_EXEC)]        = "ex",
 578                 [ilog2(VM_SHARED)]      = "sh",
 579                 [ilog2(VM_MAYREAD)]     = "mr",
 580                 [ilog2(VM_MAYWRITE)]    = "mw",
 581                 [ilog2(VM_MAYEXEC)]     = "me",
 582                 [ilog2(VM_MAYSHARE)]    = "ms",
 583                 [ilog2(VM_GROWSDOWN)]   = "gd",
 584                 [ilog2(VM_PFNMAP)]      = "pf",
 585                 [ilog2(VM_DENYWRITE)]   = "dw",
 586 #ifdef CONFIG_X86_INTEL_MPX
 587                 [ilog2(VM_MPX)]         = "mp",
 588 #endif
 589                 [ilog2(VM_LOCKED)]      = "lo",
 590                 [ilog2(VM_IO)]          = "io",
 591                 [ilog2(VM_SEQ_READ)]    = "sr",
 592                 [ilog2(VM_RAND_READ)]   = "rr",
 593                 [ilog2(VM_DONTCOPY)]    = "dc",
 594                 [ilog2(VM_DONTEXPAND)]  = "de",
 595                 [ilog2(VM_ACCOUNT)]     = "ac",
 596                 [ilog2(VM_NORESERVE)]   = "nr",
 597                 [ilog2(VM_HUGETLB)]     = "ht",
 598                 [ilog2(VM_ARCH_1)]      = "ar",
 599                 [ilog2(VM_DONTDUMP)]    = "dd",
 600 #ifdef CONFIG_MEM_SOFT_DIRTY
 601                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 602 #endif
 603                 [ilog2(VM_MIXEDMAP)]    = "mm",
 604                 [ilog2(VM_HUGEPAGE)]    = "hg",
 605                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 606                 [ilog2(VM_MERGEABLE)]   = "mg",
 607                 [ilog2(VM_UFFD_MISSING)]= "um",
 608                 [ilog2(VM_UFFD_WP)]     = "uw",
 609         };
 610         size_t i;
 611
 612         seq_puts(m, "VmFlags: ");
 613         for (i = 0; i < BITS_PER_LONG; i++) {
 614                 if (vma->vm_flags & (1UL << i)) {
 615                         seq_printf(m, "%c%c ",
 616                                    mnemonics[i][0], mnemonics[i][1]);
 617                 }
 618         }
 619         seq_putc(m, '\n');
 620 }
 621
 622 #ifdef CONFIG_HUGETLB_PAGE
 623 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 624                                  unsigned long addr, unsigned long end,
 625                                  struct mm_walk *walk)
 626 {
 627         struct mem_size_stats *mss = walk->private;
 628         struct vm_area_struct *vma = walk->vma;
 629         struct page *page = NULL;
 630
 631         if (pte_present(*pte)) {
 632                 page = vm_normal_page(vma, addr, *pte);
 633         } else if (is_swap_pte(*pte)) {
 634                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 635
 636                 if (is_migration_entry(swpent))
 637                         page = migration_entry_to_page(swpent);
 638         }
 639         if (page) {
 640                 int mapcount = page_mapcount(page);
 641
 642                 if (mapcount >= 2)
 643                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 644                 else
 645                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 646         }
 647         return 0;
 648 }
 649 #endif /* HUGETLB_PAGE */
 650
 651 static int show_smap(struct seq_file *m, void *v, int is_pid)
 652 {
 653         struct vm_area_struct *vma = v;
 654         struct mem_size_stats mss;
 655         struct mm_walk smaps_walk = {
 656                 .pmd_entry = smaps_pte_range,
 657 #ifdef CONFIG_HUGETLB_PAGE
 658                 .hugetlb_entry = smaps_hugetlb_range,
 659 #endif
 660                 .mm = vma->vm_mm,
 661                 .private = &mss,
 662         };
 663
 664         memset(&mss, 0, sizeof mss);
 665         /* mmap_sem is held in m_start */
 666         walk_page_vma(vma, &smaps_walk);
 667
 668         show_map_vma(m, vma, is_pid);
 669
 670         seq_printf(m,
 671                    "Size:           %8lu kB\n"
 672                    "Rss:            %8lu kB\n"
 673                    "Pss:            %8lu kB\n"
 674                    "Shared_Clean:   %8lu kB\n"
 675                    "Shared_Dirty:   %8lu kB\n"
 676                    "Private_Clean:  %8lu kB\n"
 677                    "Private_Dirty:  %8lu kB\n"
 678                    "Referenced:     %8lu kB\n"
 679                    "Anonymous:      %8lu kB\n"
 680                    "AnonHugePages:  %8lu kB\n"
 681                    "Shared_Hugetlb: %8lu kB\n"
 682                    "Private_Hugetlb: %7lu kB\n"
 683                    "Swap:           %8lu kB\n"
 684                    "SwapPss:        %8lu kB\n"
 685                    "KernelPageSize: %8lu kB\n"
 686                    "MMUPageSize:    %8lu kB\n"
 687                    "Locked:         %8lu kB\n",
 688                    (vma->vm_end - vma->vm_start) >> 10,
 689                    mss.resident >> 10,
 690                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 691                    mss.shared_clean  >> 10,
 692                    mss.shared_dirty  >> 10,
 693                    mss.private_clean >> 10,
 694                    mss.private_dirty >> 10,
 695                    mss.referenced >> 10,
 696                    mss.anonymous >> 10,
 697                    mss.anonymous_thp >> 10,
 698                    mss.shared_hugetlb >> 10,
 699                    mss.private_hugetlb >> 10,
 700                    mss.swap >> 10,
 701                    (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
 702                    vma_kernel_pagesize(vma) >> 10,
 703                    vma_mmu_pagesize(vma) >> 10,
 704                    (vma->vm_flags & VM_LOCKED) ?
 705                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 706
 707         show_smap_vma_flags(m, vma);
 708         m_cache_vma(m, vma);
 709         return 0;
 710 }
 711
 712 static int show_pid_smap(struct seq_file *m, void *v)
 713 {
 714         return show_smap(m, v, 1);
 715 }
 716
 717 static int show_tid_smap(struct seq_file *m, void *v)
 718 {
 719         return show_smap(m, v, 0);
 720 }
 721
 722 static const struct seq_operations proc_pid_smaps_op = {
 723         .start  = m_start,
 724         .next   = m_next,
 725         .stop   = m_stop,
 726         .show   = show_pid_smap
 727 };
 728
 729 static const struct seq_operations proc_tid_smaps_op = {
 730         .start  = m_start,
 731         .next   = m_next,
 732         .stop   = m_stop,
 733         .show   = show_tid_smap
 734 };
 735
 736 static int pid_smaps_open(struct inode *inode, struct file *file)
 737 {
 738         return do_maps_open(inode, file, &proc_pid_smaps_op);
 739 }
 740
 741 static int tid_smaps_open(struct inode *inode, struct file *file)
 742 {
 743         return do_maps_open(inode, file, &proc_tid_smaps_op);
 744 }
 745
 746 const struct file_operations proc_pid_smaps_operations = {
 747         .open           = pid_smaps_open,
 748         .read           = seq_read,
 749         .llseek         = seq_lseek,
 750         .release        = proc_map_release,
 751 };
 752
 753 const struct file_operations proc_tid_smaps_operations = {
 754         .open           = tid_smaps_open,
 755         .read           = seq_read,
 756         .llseek         = seq_lseek,
 757         .release        = proc_map_release,
 758 };
 759
 760 enum clear_refs_types {
 761         CLEAR_REFS_ALL = 1,
 762         CLEAR_REFS_ANON,
 763         CLEAR_REFS_MAPPED,
 764         CLEAR_REFS_SOFT_DIRTY,
 765         CLEAR_REFS_MM_HIWATER_RSS,
 766         CLEAR_REFS_LAST,
 767 };
 768
 769 struct clear_refs_private {
 770         enum clear_refs_types type;
 771 };
 772
 773 #ifdef CONFIG_MEM_SOFT_DIRTY
 774 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 775                 unsigned long addr, pte_t *pte)
 776 {
 777         /*
 778          * The soft-dirty tracker uses #PF-s to catch writes
 779          * to pages, so write-protect the pte as well. See the
 780          * Documentation/vm/soft-dirty.txt for full description
 781          * of how soft-dirty works.
 782          */
 783         pte_t ptent = *pte;
 784
 785         if (pte_present(ptent)) {
 786                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 787                 ptent = pte_wrprotect(ptent);
 788                 ptent = pte_clear_soft_dirty(ptent);
 789                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 790         } else if (is_swap_pte(ptent)) {
 791                 ptent = pte_swp_clear_soft_dirty(ptent);
 792                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 793         }
 794 }
 795 #else
 796 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 797                 unsigned long addr, pte_t *pte)
 798 {
 799 }
 800 #endif
 801
 802 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 803 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 804                 unsigned long addr, pmd_t *pmdp)
 805 {
 806         pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
 807
 808         pmd = pmd_wrprotect(pmd);
 809         pmd = pmd_clear_soft_dirty(pmd);
 810
 811         if (vma->vm_flags & VM_SOFTDIRTY)
 812                 vma->vm_flags &= ~VM_SOFTDIRTY;
 813
 814         set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 815 }
 816 #else
 817 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 818                 unsigned long addr, pmd_t *pmdp)
 819 {
 820 }
 821 #endif
 822
 823 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 824                                 unsigned long end, struct mm_walk *walk)
 825 {
 826         struct clear_refs_private *cp = walk->private;
 827         struct vm_area_struct *vma = walk->vma;
 828         pte_t *pte, ptent;
 829         spinlock_t *ptl;
 830         struct page *page;
 831
 832         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
 833                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 834                         clear_soft_dirty_pmd(vma, addr, pmd);
 835                         goto out;
 836                 }
 837
 838                 page = pmd_page(*pmd);
 839
 840                 /* Clear accessed and referenced bits. */
 841                 pmdp_test_and_clear_young(vma, addr, pmd);
 842                 test_and_clear_page_young(page);
 843                 ClearPageReferenced(page);
 844 out:
 845                 spin_unlock(ptl);
 846                 return 0;
 847         }
 848
 849         if (pmd_trans_unstable(pmd))
 850                 return 0;
 851
 852         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 853         for (; addr != end; pte++, addr += PAGE_SIZE) {
 854                 ptent = *pte;
 855
 856                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 857                         clear_soft_dirty(vma, addr, pte);
 858                         continue;
 859                 }
 860
 861                 if (!pte_present(ptent))
 862                         continue;
 863
 864                 page = vm_normal_page(vma, addr, ptent);
 865                 if (!page)
 866                         continue;
 867
 868                 /* Clear accessed and referenced bits. */
 869                 ptep_test_and_clear_young(vma, addr, pte);
 870                 test_and_clear_page_young(page);
 871                 ClearPageReferenced(page);
 872         }
 873         pte_unmap_unlock(pte - 1, ptl);
 874         cond_resched();
 875         return 0;
 876 }
 877
 878 static int clear_refs_test_walk(unsigned long start, unsigned long end,
 879                                 struct mm_walk *walk)
 880 {
 881         struct clear_refs_private *cp = walk->private;
 882         struct vm_area_struct *vma = walk->vma;
 883
 884         if (vma->vm_flags & VM_PFNMAP)
 885                 return 1;
 886
 887         /*
 888          * Writing 1 to /proc/pid/clear_refs affects all pages.
 889          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
 890          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
 891          * Writing 4 to /proc/pid/clear_refs affects all pages.
 892          */
 893         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
 894                 return 1;
 895         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
 896                 return 1;
 897         return 0;
 898 }
 899
 900 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 901                                 size_t count, loff_t *ppos)
 902 {
 903         struct task_struct *task;
 904         char buffer[PROC_NUMBUF];
 905         struct mm_struct *mm;
 906         struct vm_area_struct *vma;
 907         enum clear_refs_types type;
 908         int itype;
 909         int rv;
 910
 911         memset(buffer, 0, sizeof(buffer));
 912         if (count > sizeof(buffer) - 1)
 913                 count = sizeof(buffer) - 1;
 914         if (copy_from_user(buffer, buf, count))
 915                 return -EFAULT;
 916         rv = kstrtoint(strstrip(buffer), 10, &itype);
 917         if (rv < 0)
 918                 return rv;
 919         type = (enum clear_refs_types)itype;
 920         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
 921                 return -EINVAL;
 922
 923         task = get_proc_task(file_inode(file));
 924         if (!task)
 925                 return -ESRCH;
 926         mm = get_task_mm(task);
 927         if (mm) {
 928                 struct clear_refs_private cp = {
 929                         .type = type,
 930                 };
 931                 struct mm_walk clear_refs_walk = {
 932                         .pmd_entry = clear_refs_pte_range,
 933                         .test_walk = clear_refs_test_walk,
 934                         .mm = mm,
 935                         .private = &cp,
 936                 };
 937
 938                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
 939                         /*
 940                          * Writing 5 to /proc/pid/clear_refs resets the peak
 941                          * resident set size to this mm's current rss value.
 942                          */
 943                         down_write(&mm->mmap_sem);
 944                         reset_mm_hiwater_rss(mm);
 945                         up_write(&mm->mmap_sem);
 946                         goto out_mm;
 947                 }
 948
 949                 down_read(&mm->mmap_sem);
 950                 if (type == CLEAR_REFS_SOFT_DIRTY) {
 951                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
 952                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
 953                                         continue;
 954                                 up_read(&mm->mmap_sem);
 955                                 down_write(&mm->mmap_sem);
 956                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
 957                                         vma->vm_flags &= ~VM_SOFTDIRTY;
 958                                         vma_set_page_prot(vma);
 959                                 }
 960                                 downgrade_write(&mm->mmap_sem);
 961                                 break;
 962                         }
 963                         mmu_notifier_invalidate_range_start(mm, 0, -1);
 964                 }
 965                 walk_page_range(0, ~0UL, &clear_refs_walk);
 966                 if (type == CLEAR_REFS_SOFT_DIRTY)
 967                         mmu_notifier_invalidate_range_end(mm, 0, -1);
 968                 flush_tlb_mm(mm);
 969                 up_read(&mm->mmap_sem);
 970 out_mm:
 971                 mmput(mm);
 972         }
 973         put_task_struct(task);
 974
 975         return count;
 976 }
 977
 978 const struct file_operations proc_clear_refs_operations = {
 979         .write          = clear_refs_write,
 980         .llseek         = noop_llseek,
 981 };
 982
 983 typedef struct {
 984         u64 pme;
 985 } pagemap_entry_t;
 986
 987 struct pagemapread {
 988         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
 989         pagemap_entry_t *buffer;
 990         bool show_pfn;
 991 };
 992
 993 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
 994 #define PAGEMAP_WALK_MASK       (PMD_MASK)
 995
 996 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
 997 #define PM_PFRAME_BITS          55
 998 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
 999 #define PM_SOFT_DIRTY           BIT_ULL(55)
1000 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1001 #define PM_FILE                 BIT_ULL(61)
1002 #define PM_SWAP                 BIT_ULL(62)
1003 #define PM_PRESENT              BIT_ULL(63)
1004
1005 #define PM_END_OF_BUFFER    1
1006
1007 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1008 {
1009         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1010 }
1011
1012 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1013                           struct pagemapread *pm)
1014 {
1015         pm->buffer[pm->pos++] = *pme;
1016         if (pm->pos >= pm->len)
1017                 return PM_END_OF_BUFFER;
1018         return 0;
1019 }
1020
1021 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1022                                 struct mm_walk *walk)
1023 {
1024         struct pagemapread *pm = walk->private;
1025         unsigned long addr = start;
1026         int err = 0;
1027
1028         while (addr < end) {
1029                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1030                 pagemap_entry_t pme = make_pme(0, 0);
1031                 /* End of address space hole, which we mark as non-present. */
1032                 unsigned long hole_end;
1033
1034                 if (vma)
1035                         hole_end = min(end, vma->vm_start);
1036                 else
1037                         hole_end = end;
1038
1039                 for (; addr < hole_end; addr += PAGE_SIZE) {
1040                         err = add_to_pagemap(addr, &pme, pm);
1041                         if (err)
1042                                 goto out;
1043                 }
1044
1045                 if (!vma)
1046                         break;
1047
1048                 /* Addresses in the VMA. */
1049                 if (vma->vm_flags & VM_SOFTDIRTY)
1050                         pme = make_pme(0, PM_SOFT_DIRTY);
1051                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1052                         err = add_to_pagemap(addr, &pme, pm);
1053                         if (err)
1054                                 goto out;
1055                 }
1056         }
1057 out:
1058         return err;
1059 }
1060
1061 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1062                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1063 {
1064         u64 frame = 0, flags = 0;
1065         struct page *page = NULL;
1066
1067         if (pte_present(pte)) {
1068                 if (pm->show_pfn)
1069                         frame = pte_pfn(pte);
1070                 flags |= PM_PRESENT;
1071                 page = vm_normal_page(vma, addr, pte);
1072                 if (pte_soft_dirty(pte))
1073                         flags |= PM_SOFT_DIRTY;
1074         } else if (is_swap_pte(pte)) {
1075                 swp_entry_t entry;
1076                 if (pte_swp_soft_dirty(pte))
1077                         flags |= PM_SOFT_DIRTY;
1078                 entry = pte_to_swp_entry(pte);
1079                 frame = swp_type(entry) |
1080                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1081                 flags |= PM_SWAP;
1082                 if (is_migration_entry(entry))
1083                         page = migration_entry_to_page(entry);
1084         }
1085
1086         if (page && !PageAnon(page))
1087                 flags |= PM_FILE;
1088         if (page && page_mapcount(page) == 1)
1089                 flags |= PM_MMAP_EXCLUSIVE;
1090         if (vma->vm_flags & VM_SOFTDIRTY)
1091                 flags |= PM_SOFT_DIRTY;
1092
1093         return make_pme(frame, flags);
1094 }
1095
1096 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1097                              struct mm_walk *walk)
1098 {
1099         struct vm_area_struct *vma = walk->vma;
1100         struct pagemapread *pm = walk->private;
1101         spinlock_t *ptl;
1102         pte_t *pte, *orig_pte;
1103         int err = 0;
1104
1105 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1106         if (pmd_trans_huge_lock(pmdp, vma, &ptl) == 1) {
1107                 u64 flags = 0, frame = 0;
1108                 pmd_t pmd = *pmdp;
1109
1110                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1111                         flags |= PM_SOFT_DIRTY;
1112
1113                 /*
1114                  * Currently pmd for thp is always present because thp
1115                  * can not be swapped-out, migrated, or HWPOISONed
1116                  * (split in such cases instead.)
1117                  * This if-check is just to prepare for future implementation.
1118                  */
1119                 if (pmd_present(pmd)) {
1120                         struct page *page = pmd_page(pmd);
1121
1122                         if (page_mapcount(page) == 1)
1123                                 flags |= PM_MMAP_EXCLUSIVE;
1124
1125                         flags |= PM_PRESENT;
1126                         if (pm->show_pfn)
1127                                 frame = pmd_pfn(pmd) +
1128                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1129                 }
1130
1131                 for (; addr != end; addr += PAGE_SIZE) {
1132                         pagemap_entry_t pme = make_pme(frame, flags);
1133
1134                         err = add_to_pagemap(addr, &pme, pm);
1135                         if (err)
1136                                 break;
1137                         if (pm->show_pfn && (flags & PM_PRESENT))
1138                                 frame++;
1139                 }
1140                 spin_unlock(ptl);
1141                 return err;
1142         }
1143
1144         if (pmd_trans_unstable(pmdp))
1145                 return 0;
1146 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1147
1148         /*
1149          * We can assume that @vma always points to a valid one and @end never
1150          * goes beyond vma->vm_end.
1151          */
1152         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1153         for (; addr < end; pte++, addr += PAGE_SIZE) {
1154                 pagemap_entry_t pme;
1155
1156                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1157                 err = add_to_pagemap(addr, &pme, pm);
1158                 if (err)
1159                         break;
1160         }
1161         pte_unmap_unlock(orig_pte, ptl);
1162
1163         cond_resched();
1164
1165         return err;
1166 }
1167
1168 #ifdef CONFIG_HUGETLB_PAGE
1169 /* This function walks within one hugetlb entry in the single call */
1170 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1171                                  unsigned long addr, unsigned long end,
1172                                  struct mm_walk *walk)
1173 {
1174         struct pagemapread *pm = walk->private;
1175         struct vm_area_struct *vma = walk->vma;
1176         u64 flags = 0, frame = 0;
1177         int err = 0;
1178         pte_t pte;
1179
1180         if (vma->vm_flags & VM_SOFTDIRTY)
1181                 flags |= PM_SOFT_DIRTY;
1182
1183         pte = huge_ptep_get(ptep);
1184         if (pte_present(pte)) {
1185                 struct page *page = pte_page(pte);
1186
1187                 if (!PageAnon(page))
1188                         flags |= PM_FILE;
1189
1190                 if (page_mapcount(page) == 1)
1191                         flags |= PM_MMAP_EXCLUSIVE;
1192
1193                 flags |= PM_PRESENT;
1194                 if (pm->show_pfn)
1195                         frame = pte_pfn(pte) +
1196                                 ((addr & ~hmask) >> PAGE_SHIFT);
1197         }
1198
1199         for (; addr != end; addr += PAGE_SIZE) {
1200                 pagemap_entry_t pme = make_pme(frame, flags);
1201
1202                 err = add_to_pagemap(addr, &pme, pm);
1203                 if (err)
1204                         return err;
1205                 if (pm->show_pfn && (flags & PM_PRESENT))
1206                         frame++;
1207         }
1208
1209         cond_resched();
1210
1211         return err;
1212 }
1213 #endif /* HUGETLB_PAGE */
1214
1215 /*
1216  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1217  *
1218  * For each page in the address space, this file contains one 64-bit entry
1219  * consisting of the following:
1220  *
1221  * Bits 0-54  page frame number (PFN) if present
1222  * Bits 0-4   swap type if swapped
1223  * Bits 5-54  swap offset if swapped
1224  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1225  * Bit  56    page exclusively mapped
1226  * Bits 57-60 zero
1227  * Bit  61    page is file-page or shared-anon
1228  * Bit  62    page swapped
1229  * Bit  63    page present
1230  *
1231  * If the page is not present but in swap, then the PFN contains an
1232  * encoding of the swap file number and the page's offset into the
1233  * swap. Unmapped pages return a null PFN. This allows determining
1234  * precisely which pages are mapped (or in swap) and comparing mapped
1235  * pages between processes.
1236  *
1237  * Efficient users of this interface will use /proc/pid/maps to
1238  * determine which areas of memory are actually mapped and llseek to
1239  * skip over unmapped regions.
1240  */
1241 static ssize_t pagemap_read(struct file *file, char __user *buf,
1242                             size_t count, loff_t *ppos)
1243 {
1244         struct mm_struct *mm = file->private_data;
1245         struct pagemapread pm;
1246         struct mm_walk pagemap_walk = {};
1247         unsigned long src;
1248         unsigned long svpfn;
1249         unsigned long start_vaddr;
1250         unsigned long end_vaddr;
1251         int ret = 0, copied = 0;
1252
1253         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1254                 goto out;
1255
1256         ret = -EINVAL;
1257         /* file position must be aligned */
1258         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1259                 goto out_mm;
1260
1261         ret = 0;
1262         if (!count)
1263                 goto out_mm;
1264
1265         /* do not disclose physical addresses: attack vector */
1266         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1267
1268         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1269         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1270         ret = -ENOMEM;
1271         if (!pm.buffer)
1272                 goto out_mm;
1273
1274         pagemap_walk.pmd_entry = pagemap_pmd_range;
1275         pagemap_walk.pte_hole = pagemap_pte_hole;
1276 #ifdef CONFIG_HUGETLB_PAGE
1277         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1278 #endif
1279         pagemap_walk.mm = mm;
1280         pagemap_walk.private = &pm;
1281
1282         src = *ppos;
1283         svpfn = src / PM_ENTRY_BYTES;
1284         start_vaddr = svpfn << PAGE_SHIFT;
1285         end_vaddr = mm->task_size;
1286
1287         /* watch out for wraparound */
1288         if (svpfn > mm->task_size >> PAGE_SHIFT)
1289                 start_vaddr = end_vaddr;
1290
1291         /*
1292          * The odds are that this will stop walking way
1293          * before end_vaddr, because the length of the
1294          * user buffer is tracked in "pm", and the walk
1295          * will stop when we hit the end of the buffer.
1296          */
1297         ret = 0;
1298         while (count && (start_vaddr < end_vaddr)) {
1299                 int len;
1300                 unsigned long end;
1301
1302                 pm.pos = 0;
1303                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1304                 /* overflow ? */
1305                 if (end < start_vaddr || end > end_vaddr)
1306                         end = end_vaddr;
1307                 down_read(&mm->mmap_sem);
1308                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1309                 up_read(&mm->mmap_sem);
1310                 start_vaddr = end;
1311
1312                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1313                 if (copy_to_user(buf, pm.buffer, len)) {
1314                         ret = -EFAULT;
1315                         goto out_free;
1316                 }
1317                 copied += len;
1318                 buf += len;
1319                 count -= len;
1320         }
1321         *ppos += copied;
1322         if (!ret || ret == PM_END_OF_BUFFER)
1323                 ret = copied;
1324
1325 out_free:
1326         kfree(pm.buffer);
1327 out_mm:
1328         mmput(mm);
1329 out:
1330         return ret;
1331 }
1332
1333 static int pagemap_open(struct inode *inode, struct file *file)
1334 {
1335         struct mm_struct *mm;
1336
1337         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1338         if (IS_ERR(mm))
1339                 return PTR_ERR(mm);
1340         file->private_data = mm;
1341         return 0;
1342 }
1343
1344 static int pagemap_release(struct inode *inode, struct file *file)
1345 {
1346         struct mm_struct *mm = file->private_data;
1347
1348         if (mm)
1349                 mmdrop(mm);
1350         return 0;
1351 }
1352
1353 const struct file_operations proc_pagemap_operations = {
1354         .llseek         = mem_lseek, /* borrow this */
1355         .read           = pagemap_read,
1356         .open           = pagemap_open,
1357         .release        = pagemap_release,
1358 };
1359 #endif /* CONFIG_PROC_PAGE_MONITOR */
1360
1361 #ifdef CONFIG_NUMA
1362
1363 struct numa_maps {
1364         unsigned long pages;
1365         unsigned long anon;
1366         unsigned long active;
1367         unsigned long writeback;
1368         unsigned long mapcount_max;
1369         unsigned long dirty;
1370         unsigned long swapcache;
1371         unsigned long node[MAX_NUMNODES];
1372 };
1373
1374 struct numa_maps_private {
1375         struct proc_maps_private proc_maps;
1376         struct numa_maps md;
1377 };
1378
1379 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1380                         unsigned long nr_pages)
1381 {
1382         int count = page_mapcount(page);
1383
1384         md->pages += nr_pages;
1385         if (pte_dirty || PageDirty(page))
1386                 md->dirty += nr_pages;
1387
1388         if (PageSwapCache(page))
1389                 md->swapcache += nr_pages;
1390
1391         if (PageActive(page) || PageUnevictable(page))
1392                 md->active += nr_pages;
1393
1394         if (PageWriteback(page))
1395                 md->writeback += nr_pages;
1396
1397         if (PageAnon(page))
1398                 md->anon += nr_pages;
1399
1400         if (count > md->mapcount_max)
1401                 md->mapcount_max = count;
1402
1403         md->node[page_to_nid(page)] += nr_pages;
1404 }
1405
1406 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1407                 unsigned long addr)
1408 {
1409         struct page *page;
1410         int nid;
1411
1412         if (!pte_present(pte))
1413                 return NULL;
1414
1415         page = vm_normal_page(vma, addr, pte);
1416         if (!page)
1417                 return NULL;
1418
1419         if (PageReserved(page))
1420                 return NULL;
1421
1422         nid = page_to_nid(page);
1423         if (!node_isset(nid, node_states[N_MEMORY]))
1424                 return NULL;
1425
1426         return page;
1427 }
1428
1429 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1430 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1431                                               struct vm_area_struct *vma,
1432                                               unsigned long addr)
1433 {
1434         struct page *page;
1435         int nid;
1436
1437         if (!pmd_present(pmd))
1438                 return NULL;
1439
1440         page = vm_normal_page_pmd(vma, addr, pmd);
1441         if (!page)
1442                 return NULL;
1443
1444         if (PageReserved(page))
1445                 return NULL;
1446
1447         nid = page_to_nid(page);
1448         if (!node_isset(nid, node_states[N_MEMORY]))
1449                 return NULL;
1450
1451         return page;
1452 }
1453 #endif
1454
1455 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1456                 unsigned long end, struct mm_walk *walk)
1457 {
1458         struct numa_maps *md = walk->private;
1459         struct vm_area_struct *vma = walk->vma;
1460         spinlock_t *ptl;
1461         pte_t *orig_pte;
1462         pte_t *pte;
1463
1464 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1465         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1466                 struct page *page;
1467
1468                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1469                 if (page)
1470                         gather_stats(page, md, pmd_dirty(*pmd),
1471                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1472                 spin_unlock(ptl);
1473                 return 0;
1474         }
1475
1476         if (pmd_trans_unstable(pmd))
1477                 return 0;
1478 #endif
1479         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1480         do {
1481                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1482                 if (!page)
1483                         continue;
1484                 gather_stats(page, md, pte_dirty(*pte), 1);
1485
1486         } while (pte++, addr += PAGE_SIZE, addr != end);
1487         pte_unmap_unlock(orig_pte, ptl);
1488         return 0;
1489 }
1490 #ifdef CONFIG_HUGETLB_PAGE
1491 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1492                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1493 {
1494         pte_t huge_pte = huge_ptep_get(pte);
1495         struct numa_maps *md;
1496         struct page *page;
1497
1498         if (!pte_present(huge_pte))
1499                 return 0;
1500
1501         page = pte_page(huge_pte);
1502         if (!page)
1503                 return 0;
1504
1505         md = walk->private;
1506         gather_stats(page, md, pte_dirty(huge_pte), 1);
1507         return 0;
1508 }
1509
1510 #else
1511 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1512                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1513 {
1514         return 0;
1515 }
1516 #endif
1517
1518 /*
1519  * Display pages allocated per node and memory policy via /proc.
1520  */
1521 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1522 {
1523         struct numa_maps_private *numa_priv = m->private;
1524         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1525         struct vm_area_struct *vma = v;
1526         struct numa_maps *md = &numa_priv->md;
1527         struct file *file = vma->vm_file;
1528         struct mm_struct *mm = vma->vm_mm;
1529         struct mm_walk walk = {
1530                 .hugetlb_entry = gather_hugetlb_stats,
1531                 .pmd_entry = gather_pte_stats,
1532                 .private = md,
1533                 .mm = mm,
1534         };
1535         struct mempolicy *pol;
1536         char buffer[64];
1537         int nid;
1538
1539         if (!mm)
1540                 return 0;
1541
1542         /* Ensure we start with an empty set of numa_maps statistics. */
1543         memset(md, 0, sizeof(*md));
1544
1545         pol = __get_vma_policy(vma, vma->vm_start);
1546         if (pol) {
1547                 mpol_to_str(buffer, sizeof(buffer), pol);
1548                 mpol_cond_put(pol);
1549         } else {
1550                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1551         }
1552
1553         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1554
1555         if (file) {
1556                 seq_puts(m, " file=");
1557                 seq_file_path(m, file, "\n\t= ");
1558         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1559                 seq_puts(m, " heap");
1560         } else if (is_stack(proc_priv, vma, is_pid)) {
1561                 seq_puts(m, " stack");
1562         }
1563
1564         if (is_vm_hugetlb_page(vma))
1565                 seq_puts(m, " huge");
1566
1567         /* mmap_sem is held by m_start */
1568         walk_page_vma(vma, &walk);
1569
1570         if (!md->pages)
1571                 goto out;
1572
1573         if (md->anon)
1574                 seq_printf(m, " anon=%lu", md->anon);
1575
1576         if (md->dirty)
1577                 seq_printf(m, " dirty=%lu", md->dirty);
1578
1579         if (md->pages != md->anon && md->pages != md->dirty)
1580                 seq_printf(m, " mapped=%lu", md->pages);
1581
1582         if (md->mapcount_max > 1)
1583                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1584
1585         if (md->swapcache)
1586                 seq_printf(m, " swapcache=%lu", md->swapcache);
1587
1588         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1589                 seq_printf(m, " active=%lu", md->active);
1590
1591         if (md->writeback)
1592                 seq_printf(m, " writeback=%lu", md->writeback);
1593
1594         for_each_node_state(nid, N_MEMORY)
1595                 if (md->node[nid])
1596                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1597
1598         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1599 out:
1600         seq_putc(m, '\n');
1601         m_cache_vma(m, vma);
1602         return 0;
1603 }
1604
1605 static int show_pid_numa_map(struct seq_file *m, void *v)
1606 {
1607         return show_numa_map(m, v, 1);
1608 }
1609
1610 static int show_tid_numa_map(struct seq_file *m, void *v)
1611 {
1612         return show_numa_map(m, v, 0);
1613 }
1614
1615 static const struct seq_operations proc_pid_numa_maps_op = {
1616         .start  = m_start,
1617         .next   = m_next,
1618         .stop   = m_stop,
1619         .show   = show_pid_numa_map,
1620 };
1621
1622 static const struct seq_operations proc_tid_numa_maps_op = {
1623         .start  = m_start,
1624         .next   = m_next,
1625         .stop   = m_stop,
1626         .show   = show_tid_numa_map,
1627 };
1628
1629 static int numa_maps_open(struct inode *inode, struct file *file,
1630                           const struct seq_operations *ops)
1631 {
1632         return proc_maps_open(inode, file, ops,
1633                                 sizeof(struct numa_maps_private));
1634 }
1635
1636 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1637 {
1638         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1639 }
1640
1641 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1642 {
1643         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1644 }
1645
1646 const struct file_operations proc_pid_numa_maps_operations = {
1647         .open           = pid_numa_maps_open,
1648         .read           = seq_read,
1649         .llseek         = seq_lseek,
1650         .release        = proc_map_release,
1651 };
1652
1653 const struct file_operations proc_tid_numa_maps_operations = {
1654         .open           = tid_numa_maps_open,
1655         .read           = seq_read,
1656         .llseek         = seq_lseek,
1657         .release        = proc_map_release,
1658 };
1659 #endif /* CONFIG_NUMA */