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/sched/mm.h>
  15 #include <linux/swapops.h>
  16 #include <linux/mmu_notifier.h>
  17 #include <linux/page_idle.h>
  18 #include <linux/shmem_fs.h>
  19
  20 #include <asm/elf.h>
  21 #include <linux/uaccess.h>
  22 #include <asm/tlbflush.h>
  23 #include "internal.h"
  24
  25 void task_mem(struct seq_file *m, struct mm_struct *mm)
  26 {
  27         unsigned long text, lib, swap, ptes, pmds, anon, file, shmem;
  28         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  29
  30         anon = get_mm_counter(mm, MM_ANONPAGES);
  31         file = get_mm_counter(mm, MM_FILEPAGES);
  32         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  33
  34         /*
  35          * Note: to minimize their overhead, mm maintains hiwater_vm and
  36          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  37          * collector of these hiwater stats must therefore get total_vm
  38          * and rss too, which will usually be the higher.  Barriers? not
  39          * worth the effort, such snapshots can always be inconsistent.
  40          */
  41         hiwater_vm = total_vm = mm->total_vm;
  42         if (hiwater_vm < mm->hiwater_vm)
  43                 hiwater_vm = mm->hiwater_vm;
  44         hiwater_rss = total_rss = anon + file + shmem;
  45         if (hiwater_rss < mm->hiwater_rss)
  46                 hiwater_rss = mm->hiwater_rss;
  47
  48         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  49         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  50         swap = get_mm_counter(mm, MM_SWAPENTS);
  51         ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
  52         pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
  53         seq_printf(m,
  54                 "VmPeak:\t%8lu kB\n"
  55                 "VmSize:\t%8lu kB\n"
  56                 "VmLck:\t%8lu kB\n"
  57                 "VmPin:\t%8lu kB\n"
  58                 "VmHWM:\t%8lu kB\n"
  59                 "VmRSS:\t%8lu kB\n"
  60                 "RssAnon:\t%8lu kB\n"
  61                 "RssFile:\t%8lu kB\n"
  62                 "RssShmem:\t%8lu kB\n"
  63                 "VmData:\t%8lu kB\n"
  64                 "VmStk:\t%8lu kB\n"
  65                 "VmExe:\t%8lu kB\n"
  66                 "VmLib:\t%8lu kB\n"
  67                 "VmPTE:\t%8lu kB\n"
  68                 "VmPMD:\t%8lu kB\n"
  69                 "VmSwap:\t%8lu kB\n",
  70                 hiwater_vm << (PAGE_SHIFT-10),
  71                 total_vm << (PAGE_SHIFT-10),
  72                 mm->locked_vm << (PAGE_SHIFT-10),
  73                 mm->pinned_vm << (PAGE_SHIFT-10),
  74                 hiwater_rss << (PAGE_SHIFT-10),
  75                 total_rss << (PAGE_SHIFT-10),
  76                 anon << (PAGE_SHIFT-10),
  77                 file << (PAGE_SHIFT-10),
  78                 shmem << (PAGE_SHIFT-10),
  79                 mm->data_vm << (PAGE_SHIFT-10),
  80                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  81                 ptes >> 10,
  82                 pmds >> 10,
  83                 swap << (PAGE_SHIFT-10));
  84         hugetlb_report_usage(m, mm);
  85 }
  86
  87 unsigned long task_vsize(struct mm_struct *mm)
  88 {
  89         return PAGE_SIZE * mm->total_vm;
  90 }
  91
  92 unsigned long task_statm(struct mm_struct *mm,
  93                          unsigned long *shared, unsigned long *text,
  94                          unsigned long *data, unsigned long *resident)
  95 {
  96         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  97                         get_mm_counter(mm, MM_SHMEMPAGES);
  98         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  99                                                                 >> PAGE_SHIFT;
 100         *data = mm->data_vm + mm->stack_vm;
 101         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
 102         return mm->total_vm;
 103 }
 104
 105 #ifdef CONFIG_NUMA
 106 /*
 107  * Save get_task_policy() for show_numa_map().
 108  */
 109 static void hold_task_mempolicy(struct proc_maps_private *priv)
 110 {
 111         struct task_struct *task = priv->task;
 112
 113         task_lock(task);
 114         priv->task_mempolicy = get_task_policy(task);
 115         mpol_get(priv->task_mempolicy);
 116         task_unlock(task);
 117 }
 118 static void release_task_mempolicy(struct proc_maps_private *priv)
 119 {
 120         mpol_put(priv->task_mempolicy);
 121 }
 122 #else
 123 static void hold_task_mempolicy(struct proc_maps_private *priv)
 124 {
 125 }
 126 static void release_task_mempolicy(struct proc_maps_private *priv)
 127 {
 128 }
 129 #endif
 130
 131 static void vma_stop(struct proc_maps_private *priv)
 132 {
 133         struct mm_struct *mm = priv->mm;
 134
 135         release_task_mempolicy(priv);
 136         up_read(&mm->mmap_sem);
 137         mmput(mm);
 138 }
 139
 140 static struct vm_area_struct *
 141 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
 142 {
 143         if (vma == priv->tail_vma)
 144                 return NULL;
 145         return vma->vm_next ?: priv->tail_vma;
 146 }
 147
 148 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
 149 {
 150         if (m->count < m->size) /* vma is copied successfully */
 151                 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
 152 }
 153
 154 static void *m_start(struct seq_file *m, loff_t *ppos)
 155 {
 156         struct proc_maps_private *priv = m->private;
 157         unsigned long last_addr = m->version;
 158         struct mm_struct *mm;
 159         struct vm_area_struct *vma;
 160         unsigned int pos = *ppos;
 161
 162         /* See m_cache_vma(). Zero at the start or after lseek. */
 163         if (last_addr == -1UL)
 164                 return NULL;
 165
 166         priv->task = get_proc_task(priv->inode);
 167         if (!priv->task)
 168                 return ERR_PTR(-ESRCH);
 169
 170         mm = priv->mm;
 171         if (!mm || !mmget_not_zero(mm))
 172                 return NULL;
 173
 174         down_read(&mm->mmap_sem);
 175         hold_task_mempolicy(priv);
 176         priv->tail_vma = get_gate_vma(mm);
 177
 178         if (last_addr) {
 179                 vma = find_vma(mm, last_addr - 1);
 180                 if (vma && vma->vm_start <= last_addr)
 181                         vma = m_next_vma(priv, vma);
 182                 if (vma)
 183                         return vma;
 184         }
 185
 186         m->version = 0;
 187         if (pos < mm->map_count) {
 188                 for (vma = mm->mmap; pos; pos--) {
 189                         m->version = vma->vm_start;
 190                         vma = vma->vm_next;
 191                 }
 192                 return vma;
 193         }
 194
 195         /* we do not bother to update m->version in this case */
 196         if (pos == mm->map_count && priv->tail_vma)
 197                 return priv->tail_vma;
 198
 199         vma_stop(priv);
 200         return NULL;
 201 }
 202
 203 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 204 {
 205         struct proc_maps_private *priv = m->private;
 206         struct vm_area_struct *next;
 207
 208         (*pos)++;
 209         next = m_next_vma(priv, v);
 210         if (!next)
 211                 vma_stop(priv);
 212         return next;
 213 }
 214
 215 static void m_stop(struct seq_file *m, void *v)
 216 {
 217         struct proc_maps_private *priv = m->private;
 218
 219         if (!IS_ERR_OR_NULL(v))
 220                 vma_stop(priv);
 221         if (priv->task) {
 222                 put_task_struct(priv->task);
 223                 priv->task = NULL;
 224         }
 225 }
 226
 227 static int proc_maps_open(struct inode *inode, struct file *file,
 228                         const struct seq_operations *ops, int psize)
 229 {
 230         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 231
 232         if (!priv)
 233                 return -ENOMEM;
 234
 235         priv->inode = inode;
 236         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 237         if (IS_ERR(priv->mm)) {
 238                 int err = PTR_ERR(priv->mm);
 239
 240                 seq_release_private(inode, file);
 241                 return err;
 242         }
 243
 244         return 0;
 245 }
 246
 247 static int proc_map_release(struct inode *inode, struct file *file)
 248 {
 249         struct seq_file *seq = file->private_data;
 250         struct proc_maps_private *priv = seq->private;
 251
 252         if (priv->mm)
 253                 mmdrop(priv->mm);
 254
 255         return seq_release_private(inode, file);
 256 }
 257
 258 static int do_maps_open(struct inode *inode, struct file *file,
 259                         const struct seq_operations *ops)
 260 {
 261         return proc_maps_open(inode, file, ops,
 262                                 sizeof(struct proc_maps_private));
 263 }
 264
 265 /*
 266  * Indicate if the VMA is a stack for the given task; for
 267  * /proc/PID/maps that is the stack of the main task.
 268  */
 269 static int is_stack(struct proc_maps_private *priv,
 270                     struct vm_area_struct *vma)
 271 {
 272         /*
 273          * We make no effort to guess what a given thread considers to be
 274          * its "stack".  It's not even well-defined for programs written
 275          * languages like Go.
 276          */
 277         return vma->vm_start <= vma->vm_mm->start_stack &&
 278                 vma->vm_end >= vma->vm_mm->start_stack;
 279 }
 280
 281 static void
 282 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 283 {
 284         struct mm_struct *mm = vma->vm_mm;
 285         struct file *file = vma->vm_file;
 286         struct proc_maps_private *priv = m->private;
 287         vm_flags_t flags = vma->vm_flags;
 288         unsigned long ino = 0;
 289         unsigned long long pgoff = 0;
 290         unsigned long start, end;
 291         dev_t dev = 0;
 292         const char *name = NULL;
 293
 294         if (file) {
 295                 struct inode *inode = file_inode(vma->vm_file);
 296                 dev = inode->i_sb->s_dev;
 297                 ino = inode->i_ino;
 298                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 299         }
 300
 301         start = vma->vm_start;
 302         end = vma->vm_end;
 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))
 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 lazyfree;
 440         unsigned long anonymous_thp;
 441         unsigned long shmem_thp;
 442         unsigned long swap;
 443         unsigned long shared_hugetlb;
 444         unsigned long private_hugetlb;
 445         u64 pss;
 446         u64 swap_pss;
 447         bool check_shmem_swap;
 448 };
 449
 450 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 451                 bool compound, bool young, bool dirty)
 452 {
 453         int i, nr = compound ? 1 << compound_order(page) : 1;
 454         unsigned long size = nr * PAGE_SIZE;
 455
 456         if (PageAnon(page)) {
 457                 mss->anonymous += size;
 458                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 459                         mss->lazyfree += size;
 460         }
 461
 462         mss->resident += size;
 463         /* Accumulate the size in pages that have been accessed. */
 464         if (young || page_is_young(page) || PageReferenced(page))
 465                 mss->referenced += size;
 466
 467         /*
 468          * page_count(page) == 1 guarantees the page is mapped exactly once.
 469          * If any subpage of the compound page mapped with PTE it would elevate
 470          * page_count().
 471          */
 472         if (page_count(page) == 1) {
 473                 if (dirty || PageDirty(page))
 474                         mss->private_dirty += size;
 475                 else
 476                         mss->private_clean += size;
 477                 mss->pss += (u64)size << PSS_SHIFT;
 478                 return;
 479         }
 480
 481         for (i = 0; i < nr; i++, page++) {
 482                 int mapcount = page_mapcount(page);
 483
 484                 if (mapcount >= 2) {
 485                         if (dirty || PageDirty(page))
 486                                 mss->shared_dirty += PAGE_SIZE;
 487                         else
 488                                 mss->shared_clean += PAGE_SIZE;
 489                         mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
 490                 } else {
 491                         if (dirty || PageDirty(page))
 492                                 mss->private_dirty += PAGE_SIZE;
 493                         else
 494                                 mss->private_clean += PAGE_SIZE;
 495                         mss->pss += PAGE_SIZE << PSS_SHIFT;
 496                 }
 497         }
 498 }
 499
 500 #ifdef CONFIG_SHMEM
 501 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 502                 struct mm_walk *walk)
 503 {
 504         struct mem_size_stats *mss = walk->private;
 505
 506         mss->swap += shmem_partial_swap_usage(
 507                         walk->vma->vm_file->f_mapping, addr, end);
 508
 509         return 0;
 510 }
 511 #endif
 512
 513 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 514                 struct mm_walk *walk)
 515 {
 516         struct mem_size_stats *mss = walk->private;
 517         struct vm_area_struct *vma = walk->vma;
 518         struct page *page = NULL;
 519
 520         if (pte_present(*pte)) {
 521                 page = vm_normal_page(vma, addr, *pte);
 522         } else if (is_swap_pte(*pte)) {
 523                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 524
 525                 if (!non_swap_entry(swpent)) {
 526                         int mapcount;
 527
 528                         mss->swap += PAGE_SIZE;
 529                         mapcount = swp_swapcount(swpent);
 530                         if (mapcount >= 2) {
 531                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 532
 533                                 do_div(pss_delta, mapcount);
 534                                 mss->swap_pss += pss_delta;
 535                         } else {
 536                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 537                         }
 538                 } else if (is_migration_entry(swpent))
 539                         page = migration_entry_to_page(swpent);
 540         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 541                                                         && pte_none(*pte))) {
 542                 page = find_get_entry(vma->vm_file->f_mapping,
 543                                                 linear_page_index(vma, addr));
 544                 if (!page)
 545                         return;
 546
 547                 if (radix_tree_exceptional_entry(page))
 548                         mss->swap += PAGE_SIZE;
 549                 else
 550                         put_page(page);
 551
 552                 return;
 553         }
 554
 555         if (!page)
 556                 return;
 557
 558         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
 559 }
 560
 561 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 562 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 563                 struct mm_walk *walk)
 564 {
 565         struct mem_size_stats *mss = walk->private;
 566         struct vm_area_struct *vma = walk->vma;
 567         struct page *page;
 568
 569         /* FOLL_DUMP will return -EFAULT on huge zero page */
 570         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 571         if (IS_ERR_OR_NULL(page))
 572                 return;
 573         if (PageAnon(page))
 574                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 575         else if (PageSwapBacked(page))
 576                 mss->shmem_thp += HPAGE_PMD_SIZE;
 577         else if (is_zone_device_page(page))
 578                 /* pass */;
 579         else
 580                 VM_BUG_ON_PAGE(1, page);
 581         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
 582 }
 583 #else
 584 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 585                 struct mm_walk *walk)
 586 {
 587 }
 588 #endif
 589
 590 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 591                            struct mm_walk *walk)
 592 {
 593         struct vm_area_struct *vma = walk->vma;
 594         pte_t *pte;
 595         spinlock_t *ptl;
 596
 597         ptl = pmd_trans_huge_lock(pmd, vma);
 598         if (ptl) {
 599                 smaps_pmd_entry(pmd, addr, walk);
 600                 spin_unlock(ptl);
 601                 return 0;
 602         }
 603
 604         if (pmd_trans_unstable(pmd))
 605                 return 0;
 606         /*
 607          * The mmap_sem held all the way back in m_start() is what
 608          * keeps khugepaged out of here and from collapsing things
 609          * in here.
 610          */
 611         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 612         for (; addr != end; pte++, addr += PAGE_SIZE)
 613                 smaps_pte_entry(pte, addr, walk);
 614         pte_unmap_unlock(pte - 1, ptl);
 615         cond_resched();
 616         return 0;
 617 }
 618
 619 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 620 {
 621         /*
 622          * Don't forget to update Documentation/ on changes.
 623          */
 624         static const char mnemonics[BITS_PER_LONG][2] = {
 625                 /*
 626                  * In case if we meet a flag we don't know about.
 627                  */
 628                 [0 ... (BITS_PER_LONG-1)] = "??",
 629
 630                 [ilog2(VM_READ)]        = "rd",
 631                 [ilog2(VM_WRITE)]       = "wr",
 632                 [ilog2(VM_EXEC)]        = "ex",
 633                 [ilog2(VM_SHARED)]      = "sh",
 634                 [ilog2(VM_MAYREAD)]     = "mr",
 635                 [ilog2(VM_MAYWRITE)]    = "mw",
 636                 [ilog2(VM_MAYEXEC)]     = "me",
 637                 [ilog2(VM_MAYSHARE)]    = "ms",
 638                 [ilog2(VM_GROWSDOWN)]   = "gd",
 639                 [ilog2(VM_PFNMAP)]      = "pf",
 640                 [ilog2(VM_DENYWRITE)]   = "dw",
 641 #ifdef CONFIG_X86_INTEL_MPX
 642                 [ilog2(VM_MPX)]         = "mp",
 643 #endif
 644                 [ilog2(VM_LOCKED)]      = "lo",
 645                 [ilog2(VM_IO)]          = "io",
 646                 [ilog2(VM_SEQ_READ)]    = "sr",
 647                 [ilog2(VM_RAND_READ)]   = "rr",
 648                 [ilog2(VM_DONTCOPY)]    = "dc",
 649                 [ilog2(VM_DONTEXPAND)]  = "de",
 650                 [ilog2(VM_ACCOUNT)]     = "ac",
 651                 [ilog2(VM_NORESERVE)]   = "nr",
 652                 [ilog2(VM_HUGETLB)]     = "ht",
 653                 [ilog2(VM_ARCH_1)]      = "ar",
 654                 [ilog2(VM_DONTDUMP)]    = "dd",
 655 #ifdef CONFIG_MEM_SOFT_DIRTY
 656                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 657 #endif
 658                 [ilog2(VM_MIXEDMAP)]    = "mm",
 659                 [ilog2(VM_HUGEPAGE)]    = "hg",
 660                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 661                 [ilog2(VM_MERGEABLE)]   = "mg",
 662                 [ilog2(VM_UFFD_MISSING)]= "um",
 663                 [ilog2(VM_UFFD_WP)]     = "uw",
 664 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 665                 /* These come out via ProtectionKey: */
 666                 [ilog2(VM_PKEY_BIT0)]   = "",
 667                 [ilog2(VM_PKEY_BIT1)]   = "",
 668                 [ilog2(VM_PKEY_BIT2)]   = "",
 669                 [ilog2(VM_PKEY_BIT3)]   = "",
 670 #endif
 671         };
 672         size_t i;
 673
 674         seq_puts(m, "VmFlags: ");
 675         for (i = 0; i < BITS_PER_LONG; i++) {
 676                 if (!mnemonics[i][0])
 677                         continue;
 678                 if (vma->vm_flags & (1UL << i)) {
 679                         seq_printf(m, "%c%c ",
 680                                    mnemonics[i][0], mnemonics[i][1]);
 681                 }
 682         }
 683         seq_putc(m, '\n');
 684 }
 685
 686 #ifdef CONFIG_HUGETLB_PAGE
 687 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 688                                  unsigned long addr, unsigned long end,
 689                                  struct mm_walk *walk)
 690 {
 691         struct mem_size_stats *mss = walk->private;
 692         struct vm_area_struct *vma = walk->vma;
 693         struct page *page = NULL;
 694
 695         if (pte_present(*pte)) {
 696                 page = vm_normal_page(vma, addr, *pte);
 697         } else if (is_swap_pte(*pte)) {
 698                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 699
 700                 if (is_migration_entry(swpent))
 701                         page = migration_entry_to_page(swpent);
 702         }
 703         if (page) {
 704                 int mapcount = page_mapcount(page);
 705
 706                 if (mapcount >= 2)
 707                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 708                 else
 709                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 710         }
 711         return 0;
 712 }
 713 #endif /* HUGETLB_PAGE */
 714
 715 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
 716 {
 717 }
 718
 719 static int show_smap(struct seq_file *m, void *v, int is_pid)
 720 {
 721         struct vm_area_struct *vma = v;
 722         struct mem_size_stats mss;
 723         struct mm_walk smaps_walk = {
 724                 .pmd_entry = smaps_pte_range,
 725 #ifdef CONFIG_HUGETLB_PAGE
 726                 .hugetlb_entry = smaps_hugetlb_range,
 727 #endif
 728                 .mm = vma->vm_mm,
 729                 .private = &mss,
 730         };
 731
 732         memset(&mss, 0, sizeof mss);
 733
 734 #ifdef CONFIG_SHMEM
 735         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 736                 /*
 737                  * For shared or readonly shmem mappings we know that all
 738                  * swapped out pages belong to the shmem object, and we can
 739                  * obtain the swap value much more efficiently. For private
 740                  * writable mappings, we might have COW pages that are
 741                  * not affected by the parent swapped out pages of the shmem
 742                  * object, so we have to distinguish them during the page walk.
 743                  * Unless we know that the shmem object (or the part mapped by
 744                  * our VMA) has no swapped out pages at all.
 745                  */
 746                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 747
 748                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 749                                         !(vma->vm_flags & VM_WRITE)) {
 750                         mss.swap = shmem_swapped;
 751                 } else {
 752                         mss.check_shmem_swap = true;
 753                         smaps_walk.pte_hole = smaps_pte_hole;
 754                 }
 755         }
 756 #endif
 757
 758         /* mmap_sem is held in m_start */
 759         walk_page_vma(vma, &smaps_walk);
 760
 761         show_map_vma(m, vma, is_pid);
 762
 763         seq_printf(m,
 764                    "Size:           %8lu kB\n"
 765                    "Rss:            %8lu kB\n"
 766                    "Pss:            %8lu kB\n"
 767                    "Shared_Clean:   %8lu kB\n"
 768                    "Shared_Dirty:   %8lu kB\n"
 769                    "Private_Clean:  %8lu kB\n"
 770                    "Private_Dirty:  %8lu kB\n"
 771                    "Referenced:     %8lu kB\n"
 772                    "Anonymous:      %8lu kB\n"
 773                    "LazyFree:       %8lu kB\n"
 774                    "AnonHugePages:  %8lu kB\n"
 775                    "ShmemPmdMapped: %8lu kB\n"
 776                    "Shared_Hugetlb: %8lu kB\n"
 777                    "Private_Hugetlb: %7lu kB\n"
 778                    "Swap:           %8lu kB\n"
 779                    "SwapPss:        %8lu kB\n"
 780                    "KernelPageSize: %8lu kB\n"
 781                    "MMUPageSize:    %8lu kB\n"
 782                    "Locked:         %8lu kB\n",
 783                    (vma->vm_end - vma->vm_start) >> 10,
 784                    mss.resident >> 10,
 785                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 786                    mss.shared_clean  >> 10,
 787                    mss.shared_dirty  >> 10,
 788                    mss.private_clean >> 10,
 789                    mss.private_dirty >> 10,
 790                    mss.referenced >> 10,
 791                    mss.anonymous >> 10,
 792                    mss.lazyfree >> 10,
 793                    mss.anonymous_thp >> 10,
 794                    mss.shmem_thp >> 10,
 795                    mss.shared_hugetlb >> 10,
 796                    mss.private_hugetlb >> 10,
 797                    mss.swap >> 10,
 798                    (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
 799                    vma_kernel_pagesize(vma) >> 10,
 800                    vma_mmu_pagesize(vma) >> 10,
 801                    (vma->vm_flags & VM_LOCKED) ?
 802                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 803
 804         arch_show_smap(m, vma);
 805         show_smap_vma_flags(m, vma);
 806         m_cache_vma(m, vma);
 807         return 0;
 808 }
 809
 810 static int show_pid_smap(struct seq_file *m, void *v)
 811 {
 812         return show_smap(m, v, 1);
 813 }
 814
 815 static int show_tid_smap(struct seq_file *m, void *v)
 816 {
 817         return show_smap(m, v, 0);
 818 }
 819
 820 static const struct seq_operations proc_pid_smaps_op = {
 821         .start  = m_start,
 822         .next   = m_next,
 823         .stop   = m_stop,
 824         .show   = show_pid_smap
 825 };
 826
 827 static const struct seq_operations proc_tid_smaps_op = {
 828         .start  = m_start,
 829         .next   = m_next,
 830         .stop   = m_stop,
 831         .show   = show_tid_smap
 832 };
 833
 834 static int pid_smaps_open(struct inode *inode, struct file *file)
 835 {
 836         return do_maps_open(inode, file, &proc_pid_smaps_op);
 837 }
 838
 839 static int tid_smaps_open(struct inode *inode, struct file *file)
 840 {
 841         return do_maps_open(inode, file, &proc_tid_smaps_op);
 842 }
 843
 844 const struct file_operations proc_pid_smaps_operations = {
 845         .open           = pid_smaps_open,
 846         .read           = seq_read,
 847         .llseek         = seq_lseek,
 848         .release        = proc_map_release,
 849 };
 850
 851 const struct file_operations proc_tid_smaps_operations = {
 852         .open           = tid_smaps_open,
 853         .read           = seq_read,
 854         .llseek         = seq_lseek,
 855         .release        = proc_map_release,
 856 };
 857
 858 enum clear_refs_types {
 859         CLEAR_REFS_ALL = 1,
 860         CLEAR_REFS_ANON,
 861         CLEAR_REFS_MAPPED,
 862         CLEAR_REFS_SOFT_DIRTY,
 863         CLEAR_REFS_MM_HIWATER_RSS,
 864         CLEAR_REFS_LAST,
 865 };
 866
 867 struct clear_refs_private {
 868         enum clear_refs_types type;
 869 };
 870
 871 #ifdef CONFIG_MEM_SOFT_DIRTY
 872 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 873                 unsigned long addr, pte_t *pte)
 874 {
 875         /*
 876          * The soft-dirty tracker uses #PF-s to catch writes
 877          * to pages, so write-protect the pte as well. See the
 878          * Documentation/vm/soft-dirty.txt for full description
 879          * of how soft-dirty works.
 880          */
 881         pte_t ptent = *pte;
 882
 883         if (pte_present(ptent)) {
 884                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 885                 ptent = pte_wrprotect(ptent);
 886                 ptent = pte_clear_soft_dirty(ptent);
 887                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 888         } else if (is_swap_pte(ptent)) {
 889                 ptent = pte_swp_clear_soft_dirty(ptent);
 890                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 891         }
 892 }
 893 #else
 894 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 895                 unsigned long addr, pte_t *pte)
 896 {
 897 }
 898 #endif
 899
 900 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 901 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 902                 unsigned long addr, pmd_t *pmdp)
 903 {
 904         pmd_t pmd = *pmdp;
 905
 906         /* See comment in change_huge_pmd() */
 907         pmdp_invalidate(vma, addr, pmdp);
 908         if (pmd_dirty(*pmdp))
 909                 pmd = pmd_mkdirty(pmd);
 910         if (pmd_young(*pmdp))
 911                 pmd = pmd_mkyoung(pmd);
 912
 913         pmd = pmd_wrprotect(pmd);
 914         pmd = pmd_clear_soft_dirty(pmd);
 915
 916         set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 917 }
 918 #else
 919 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 920                 unsigned long addr, pmd_t *pmdp)
 921 {
 922 }
 923 #endif
 924
 925 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 926                                 unsigned long end, struct mm_walk *walk)
 927 {
 928         struct clear_refs_private *cp = walk->private;
 929         struct vm_area_struct *vma = walk->vma;
 930         pte_t *pte, ptent;
 931         spinlock_t *ptl;
 932         struct page *page;
 933
 934         ptl = pmd_trans_huge_lock(pmd, vma);
 935         if (ptl) {
 936                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 937                         clear_soft_dirty_pmd(vma, addr, pmd);
 938                         goto out;
 939                 }
 940
 941                 page = pmd_page(*pmd);
 942
 943                 /* Clear accessed and referenced bits. */
 944                 pmdp_test_and_clear_young(vma, addr, pmd);
 945                 test_and_clear_page_young(page);
 946                 ClearPageReferenced(page);
 947 out:
 948                 spin_unlock(ptl);
 949                 return 0;
 950         }
 951
 952         if (pmd_trans_unstable(pmd))
 953                 return 0;
 954
 955         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 956         for (; addr != end; pte++, addr += PAGE_SIZE) {
 957                 ptent = *pte;
 958
 959                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 960                         clear_soft_dirty(vma, addr, pte);
 961                         continue;
 962                 }
 963
 964                 if (!pte_present(ptent))
 965                         continue;
 966
 967                 page = vm_normal_page(vma, addr, ptent);
 968                 if (!page)
 969                         continue;
 970
 971                 /* Clear accessed and referenced bits. */
 972                 ptep_test_and_clear_young(vma, addr, pte);
 973                 test_and_clear_page_young(page);
 974                 ClearPageReferenced(page);
 975         }
 976         pte_unmap_unlock(pte - 1, ptl);
 977         cond_resched();
 978         return 0;
 979 }
 980
 981 static int clear_refs_test_walk(unsigned long start, unsigned long end,
 982                                 struct mm_walk *walk)
 983 {
 984         struct clear_refs_private *cp = walk->private;
 985         struct vm_area_struct *vma = walk->vma;
 986
 987         if (vma->vm_flags & VM_PFNMAP)
 988                 return 1;
 989
 990         /*
 991          * Writing 1 to /proc/pid/clear_refs affects all pages.
 992          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
 993          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
 994          * Writing 4 to /proc/pid/clear_refs affects all pages.
 995          */
 996         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
 997                 return 1;
 998         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
 999                 return 1;
1000         return 0;
1001 }
1002
1003 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1004                                 size_t count, loff_t *ppos)
1005 {
1006         struct task_struct *task;
1007         char buffer[PROC_NUMBUF];
1008         struct mm_struct *mm;
1009         struct vm_area_struct *vma;
1010         enum clear_refs_types type;
1011         int itype;
1012         int rv;
1013
1014         memset(buffer, 0, sizeof(buffer));
1015         if (count > sizeof(buffer) - 1)
1016                 count = sizeof(buffer) - 1;
1017         if (copy_from_user(buffer, buf, count))
1018                 return -EFAULT;
1019         rv = kstrtoint(strstrip(buffer), 10, &itype);
1020         if (rv < 0)
1021                 return rv;
1022         type = (enum clear_refs_types)itype;
1023         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1024                 return -EINVAL;
1025
1026         task = get_proc_task(file_inode(file));
1027         if (!task)
1028                 return -ESRCH;
1029         mm = get_task_mm(task);
1030         if (mm) {
1031                 struct clear_refs_private cp = {
1032                         .type = type,
1033                 };
1034                 struct mm_walk clear_refs_walk = {
1035                         .pmd_entry = clear_refs_pte_range,
1036                         .test_walk = clear_refs_test_walk,
1037                         .mm = mm,
1038                         .private = &cp,
1039                 };
1040
1041                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1042                         if (down_write_killable(&mm->mmap_sem)) {
1043                                 count = -EINTR;
1044                                 goto out_mm;
1045                         }
1046
1047                         /*
1048                          * Writing 5 to /proc/pid/clear_refs resets the peak
1049                          * resident set size to this mm's current rss value.
1050                          */
1051                         reset_mm_hiwater_rss(mm);
1052                         up_write(&mm->mmap_sem);
1053                         goto out_mm;
1054                 }
1055
1056                 down_read(&mm->mmap_sem);
1057                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1058                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1059                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1060                                         continue;
1061                                 up_read(&mm->mmap_sem);
1062                                 if (down_write_killable(&mm->mmap_sem)) {
1063                                         count = -EINTR;
1064                                         goto out_mm;
1065                                 }
1066                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1067                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1068                                         vma_set_page_prot(vma);
1069                                 }
1070                                 downgrade_write(&mm->mmap_sem);
1071                                 break;
1072                         }
1073                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1074                 }
1075                 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1076                 if (type == CLEAR_REFS_SOFT_DIRTY)
1077                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1078                 flush_tlb_mm(mm);
1079                 up_read(&mm->mmap_sem);
1080 out_mm:
1081                 mmput(mm);
1082         }
1083         put_task_struct(task);
1084
1085         return count;
1086 }
1087
1088 const struct file_operations proc_clear_refs_operations = {
1089         .write          = clear_refs_write,
1090         .llseek         = noop_llseek,
1091 };
1092
1093 typedef struct {
1094         u64 pme;
1095 } pagemap_entry_t;
1096
1097 struct pagemapread {
1098         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1099         pagemap_entry_t *buffer;
1100         bool show_pfn;
1101 };
1102
1103 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1104 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1105
1106 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1107 #define PM_PFRAME_BITS          55
1108 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1109 #define PM_SOFT_DIRTY           BIT_ULL(55)
1110 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1111 #define PM_FILE                 BIT_ULL(61)
1112 #define PM_SWAP                 BIT_ULL(62)
1113 #define PM_PRESENT              BIT_ULL(63)
1114
1115 #define PM_END_OF_BUFFER    1
1116
1117 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1118 {
1119         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1120 }
1121
1122 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1123                           struct pagemapread *pm)
1124 {
1125         pm->buffer[pm->pos++] = *pme;
1126         if (pm->pos >= pm->len)
1127                 return PM_END_OF_BUFFER;
1128         return 0;
1129 }
1130
1131 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1132                                 struct mm_walk *walk)
1133 {
1134         struct pagemapread *pm = walk->private;
1135         unsigned long addr = start;
1136         int err = 0;
1137
1138         while (addr < end) {
1139                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1140                 pagemap_entry_t pme = make_pme(0, 0);
1141                 /* End of address space hole, which we mark as non-present. */
1142                 unsigned long hole_end;
1143
1144                 if (vma)
1145                         hole_end = min(end, vma->vm_start);
1146                 else
1147                         hole_end = end;
1148
1149                 for (; addr < hole_end; addr += PAGE_SIZE) {
1150                         err = add_to_pagemap(addr, &pme, pm);
1151                         if (err)
1152                                 goto out;
1153                 }
1154
1155                 if (!vma)
1156                         break;
1157
1158                 /* Addresses in the VMA. */
1159                 if (vma->vm_flags & VM_SOFTDIRTY)
1160                         pme = make_pme(0, PM_SOFT_DIRTY);
1161                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1162                         err = add_to_pagemap(addr, &pme, pm);
1163                         if (err)
1164                                 goto out;
1165                 }
1166         }
1167 out:
1168         return err;
1169 }
1170
1171 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1172                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1173 {
1174         u64 frame = 0, flags = 0;
1175         struct page *page = NULL;
1176
1177         if (pte_present(pte)) {
1178                 if (pm->show_pfn)
1179                         frame = pte_pfn(pte);
1180                 flags |= PM_PRESENT;
1181                 page = vm_normal_page(vma, addr, pte);
1182                 if (pte_soft_dirty(pte))
1183                         flags |= PM_SOFT_DIRTY;
1184         } else if (is_swap_pte(pte)) {
1185                 swp_entry_t entry;
1186                 if (pte_swp_soft_dirty(pte))
1187                         flags |= PM_SOFT_DIRTY;
1188                 entry = pte_to_swp_entry(pte);
1189                 frame = swp_type(entry) |
1190                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1191                 flags |= PM_SWAP;
1192                 if (is_migration_entry(entry))
1193                         page = migration_entry_to_page(entry);
1194         }
1195
1196         if (page && !PageAnon(page))
1197                 flags |= PM_FILE;
1198         if (page && page_mapcount(page) == 1)
1199                 flags |= PM_MMAP_EXCLUSIVE;
1200         if (vma->vm_flags & VM_SOFTDIRTY)
1201                 flags |= PM_SOFT_DIRTY;
1202
1203         return make_pme(frame, flags);
1204 }
1205
1206 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1207                              struct mm_walk *walk)
1208 {
1209         struct vm_area_struct *vma = walk->vma;
1210         struct pagemapread *pm = walk->private;
1211         spinlock_t *ptl;
1212         pte_t *pte, *orig_pte;
1213         int err = 0;
1214
1215 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1216         ptl = pmd_trans_huge_lock(pmdp, vma);
1217         if (ptl) {
1218                 u64 flags = 0, frame = 0;
1219                 pmd_t pmd = *pmdp;
1220
1221                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1222                         flags |= PM_SOFT_DIRTY;
1223
1224                 /*
1225                  * Currently pmd for thp is always present because thp
1226                  * can not be swapped-out, migrated, or HWPOISONed
1227                  * (split in such cases instead.)
1228                  * This if-check is just to prepare for future implementation.
1229                  */
1230                 if (pmd_present(pmd)) {
1231                         struct page *page = pmd_page(pmd);
1232
1233                         if (page_mapcount(page) == 1)
1234                                 flags |= PM_MMAP_EXCLUSIVE;
1235
1236                         flags |= PM_PRESENT;
1237                         if (pm->show_pfn)
1238                                 frame = pmd_pfn(pmd) +
1239                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1240                 }
1241
1242                 for (; addr != end; addr += PAGE_SIZE) {
1243                         pagemap_entry_t pme = make_pme(frame, flags);
1244
1245                         err = add_to_pagemap(addr, &pme, pm);
1246                         if (err)
1247                                 break;
1248                         if (pm->show_pfn && (flags & PM_PRESENT))
1249                                 frame++;
1250                 }
1251                 spin_unlock(ptl);
1252                 return err;
1253         }
1254
1255         if (pmd_trans_unstable(pmdp))
1256                 return 0;
1257 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1258
1259         /*
1260          * We can assume that @vma always points to a valid one and @end never
1261          * goes beyond vma->vm_end.
1262          */
1263         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1264         for (; addr < end; pte++, addr += PAGE_SIZE) {
1265                 pagemap_entry_t pme;
1266
1267                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1268                 err = add_to_pagemap(addr, &pme, pm);
1269                 if (err)
1270                         break;
1271         }
1272         pte_unmap_unlock(orig_pte, ptl);
1273
1274         cond_resched();
1275
1276         return err;
1277 }
1278
1279 #ifdef CONFIG_HUGETLB_PAGE
1280 /* This function walks within one hugetlb entry in the single call */
1281 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1282                                  unsigned long addr, unsigned long end,
1283                                  struct mm_walk *walk)
1284 {
1285         struct pagemapread *pm = walk->private;
1286         struct vm_area_struct *vma = walk->vma;
1287         u64 flags = 0, frame = 0;
1288         int err = 0;
1289         pte_t pte;
1290
1291         if (vma->vm_flags & VM_SOFTDIRTY)
1292                 flags |= PM_SOFT_DIRTY;
1293
1294         pte = huge_ptep_get(ptep);
1295         if (pte_present(pte)) {
1296                 struct page *page = pte_page(pte);
1297
1298                 if (!PageAnon(page))
1299                         flags |= PM_FILE;
1300
1301                 if (page_mapcount(page) == 1)
1302                         flags |= PM_MMAP_EXCLUSIVE;
1303
1304                 flags |= PM_PRESENT;
1305                 if (pm->show_pfn)
1306                         frame = pte_pfn(pte) +
1307                                 ((addr & ~hmask) >> PAGE_SHIFT);
1308         }
1309
1310         for (; addr != end; addr += PAGE_SIZE) {
1311                 pagemap_entry_t pme = make_pme(frame, flags);
1312
1313                 err = add_to_pagemap(addr, &pme, pm);
1314                 if (err)
1315                         return err;
1316                 if (pm->show_pfn && (flags & PM_PRESENT))
1317                         frame++;
1318         }
1319
1320         cond_resched();
1321
1322         return err;
1323 }
1324 #endif /* HUGETLB_PAGE */
1325
1326 /*
1327  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1328  *
1329  * For each page in the address space, this file contains one 64-bit entry
1330  * consisting of the following:
1331  *
1332  * Bits 0-54  page frame number (PFN) if present
1333  * Bits 0-4   swap type if swapped
1334  * Bits 5-54  swap offset if swapped
1335  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1336  * Bit  56    page exclusively mapped
1337  * Bits 57-60 zero
1338  * Bit  61    page is file-page or shared-anon
1339  * Bit  62    page swapped
1340  * Bit  63    page present
1341  *
1342  * If the page is not present but in swap, then the PFN contains an
1343  * encoding of the swap file number and the page's offset into the
1344  * swap. Unmapped pages return a null PFN. This allows determining
1345  * precisely which pages are mapped (or in swap) and comparing mapped
1346  * pages between processes.
1347  *
1348  * Efficient users of this interface will use /proc/pid/maps to
1349  * determine which areas of memory are actually mapped and llseek to
1350  * skip over unmapped regions.
1351  */
1352 static ssize_t pagemap_read(struct file *file, char __user *buf,
1353                             size_t count, loff_t *ppos)
1354 {
1355         struct mm_struct *mm = file->private_data;
1356         struct pagemapread pm;
1357         struct mm_walk pagemap_walk = {};
1358         unsigned long src;
1359         unsigned long svpfn;
1360         unsigned long start_vaddr;
1361         unsigned long end_vaddr;
1362         int ret = 0, copied = 0;
1363
1364         if (!mm || !mmget_not_zero(mm))
1365                 goto out;
1366
1367         ret = -EINVAL;
1368         /* file position must be aligned */
1369         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1370                 goto out_mm;
1371
1372         ret = 0;
1373         if (!count)
1374                 goto out_mm;
1375
1376         /* do not disclose physical addresses: attack vector */
1377         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1378
1379         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1380         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1381         ret = -ENOMEM;
1382         if (!pm.buffer)
1383                 goto out_mm;
1384
1385         pagemap_walk.pmd_entry = pagemap_pmd_range;
1386         pagemap_walk.pte_hole = pagemap_pte_hole;
1387 #ifdef CONFIG_HUGETLB_PAGE
1388         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1389 #endif
1390         pagemap_walk.mm = mm;
1391         pagemap_walk.private = &pm;
1392
1393         src = *ppos;
1394         svpfn = src / PM_ENTRY_BYTES;
1395         start_vaddr = svpfn << PAGE_SHIFT;
1396         end_vaddr = mm->task_size;
1397
1398         /* watch out for wraparound */
1399         if (svpfn > mm->task_size >> PAGE_SHIFT)
1400                 start_vaddr = end_vaddr;
1401
1402         /*
1403          * The odds are that this will stop walking way
1404          * before end_vaddr, because the length of the
1405          * user buffer is tracked in "pm", and the walk
1406          * will stop when we hit the end of the buffer.
1407          */
1408         ret = 0;
1409         while (count && (start_vaddr < end_vaddr)) {
1410                 int len;
1411                 unsigned long end;
1412
1413                 pm.pos = 0;
1414                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1415                 /* overflow ? */
1416                 if (end < start_vaddr || end > end_vaddr)
1417                         end = end_vaddr;
1418                 down_read(&mm->mmap_sem);
1419                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1420                 up_read(&mm->mmap_sem);
1421                 start_vaddr = end;
1422
1423                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1424                 if (copy_to_user(buf, pm.buffer, len)) {
1425                         ret = -EFAULT;
1426                         goto out_free;
1427                 }
1428                 copied += len;
1429                 buf += len;
1430                 count -= len;
1431         }
1432         *ppos += copied;
1433         if (!ret || ret == PM_END_OF_BUFFER)
1434                 ret = copied;
1435
1436 out_free:
1437         kfree(pm.buffer);
1438 out_mm:
1439         mmput(mm);
1440 out:
1441         return ret;
1442 }
1443
1444 static int pagemap_open(struct inode *inode, struct file *file)
1445 {
1446         struct mm_struct *mm;
1447
1448         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1449         if (IS_ERR(mm))
1450                 return PTR_ERR(mm);
1451         file->private_data = mm;
1452         return 0;
1453 }
1454
1455 static int pagemap_release(struct inode *inode, struct file *file)
1456 {
1457         struct mm_struct *mm = file->private_data;
1458
1459         if (mm)
1460                 mmdrop(mm);
1461         return 0;
1462 }
1463
1464 const struct file_operations proc_pagemap_operations = {
1465         .llseek         = mem_lseek, /* borrow this */
1466         .read           = pagemap_read,
1467         .open           = pagemap_open,
1468         .release        = pagemap_release,
1469 };
1470 #endif /* CONFIG_PROC_PAGE_MONITOR */
1471
1472 #ifdef CONFIG_NUMA
1473
1474 struct numa_maps {
1475         unsigned long pages;
1476         unsigned long anon;
1477         unsigned long active;
1478         unsigned long writeback;
1479         unsigned long mapcount_max;
1480         unsigned long dirty;
1481         unsigned long swapcache;
1482         unsigned long node[MAX_NUMNODES];
1483 };
1484
1485 struct numa_maps_private {
1486         struct proc_maps_private proc_maps;
1487         struct numa_maps md;
1488 };
1489
1490 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1491                         unsigned long nr_pages)
1492 {
1493         int count = page_mapcount(page);
1494
1495         md->pages += nr_pages;
1496         if (pte_dirty || PageDirty(page))
1497                 md->dirty += nr_pages;
1498
1499         if (PageSwapCache(page))
1500                 md->swapcache += nr_pages;
1501
1502         if (PageActive(page) || PageUnevictable(page))
1503                 md->active += nr_pages;
1504
1505         if (PageWriteback(page))
1506                 md->writeback += nr_pages;
1507
1508         if (PageAnon(page))
1509                 md->anon += nr_pages;
1510
1511         if (count > md->mapcount_max)
1512                 md->mapcount_max = count;
1513
1514         md->node[page_to_nid(page)] += nr_pages;
1515 }
1516
1517 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1518                 unsigned long addr)
1519 {
1520         struct page *page;
1521         int nid;
1522
1523         if (!pte_present(pte))
1524                 return NULL;
1525
1526         page = vm_normal_page(vma, addr, pte);
1527         if (!page)
1528                 return NULL;
1529
1530         if (PageReserved(page))
1531                 return NULL;
1532
1533         nid = page_to_nid(page);
1534         if (!node_isset(nid, node_states[N_MEMORY]))
1535                 return NULL;
1536
1537         return page;
1538 }
1539
1540 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1541 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1542                                               struct vm_area_struct *vma,
1543                                               unsigned long addr)
1544 {
1545         struct page *page;
1546         int nid;
1547
1548         if (!pmd_present(pmd))
1549                 return NULL;
1550
1551         page = vm_normal_page_pmd(vma, addr, pmd);
1552         if (!page)
1553                 return NULL;
1554
1555         if (PageReserved(page))
1556                 return NULL;
1557
1558         nid = page_to_nid(page);
1559         if (!node_isset(nid, node_states[N_MEMORY]))
1560                 return NULL;
1561
1562         return page;
1563 }
1564 #endif
1565
1566 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1567                 unsigned long end, struct mm_walk *walk)
1568 {
1569         struct numa_maps *md = walk->private;
1570         struct vm_area_struct *vma = walk->vma;
1571         spinlock_t *ptl;
1572         pte_t *orig_pte;
1573         pte_t *pte;
1574
1575 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1576         ptl = pmd_trans_huge_lock(pmd, vma);
1577         if (ptl) {
1578                 struct page *page;
1579
1580                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1581                 if (page)
1582                         gather_stats(page, md, pmd_dirty(*pmd),
1583                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1584                 spin_unlock(ptl);
1585                 return 0;
1586         }
1587
1588         if (pmd_trans_unstable(pmd))
1589                 return 0;
1590 #endif
1591         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1592         do {
1593                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1594                 if (!page)
1595                         continue;
1596                 gather_stats(page, md, pte_dirty(*pte), 1);
1597
1598         } while (pte++, addr += PAGE_SIZE, addr != end);
1599         pte_unmap_unlock(orig_pte, ptl);
1600         cond_resched();
1601         return 0;
1602 }
1603 #ifdef CONFIG_HUGETLB_PAGE
1604 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1605                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1606 {
1607         pte_t huge_pte = huge_ptep_get(pte);
1608         struct numa_maps *md;
1609         struct page *page;
1610
1611         if (!pte_present(huge_pte))
1612                 return 0;
1613
1614         page = pte_page(huge_pte);
1615         if (!page)
1616                 return 0;
1617
1618         md = walk->private;
1619         gather_stats(page, md, pte_dirty(huge_pte), 1);
1620         return 0;
1621 }
1622
1623 #else
1624 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1625                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1626 {
1627         return 0;
1628 }
1629 #endif
1630
1631 /*
1632  * Display pages allocated per node and memory policy via /proc.
1633  */
1634 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1635 {
1636         struct numa_maps_private *numa_priv = m->private;
1637         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1638         struct vm_area_struct *vma = v;
1639         struct numa_maps *md = &numa_priv->md;
1640         struct file *file = vma->vm_file;
1641         struct mm_struct *mm = vma->vm_mm;
1642         struct mm_walk walk = {
1643                 .hugetlb_entry = gather_hugetlb_stats,
1644                 .pmd_entry = gather_pte_stats,
1645                 .private = md,
1646                 .mm = mm,
1647         };
1648         struct mempolicy *pol;
1649         char buffer[64];
1650         int nid;
1651
1652         if (!mm)
1653                 return 0;
1654
1655         /* Ensure we start with an empty set of numa_maps statistics. */
1656         memset(md, 0, sizeof(*md));
1657
1658         pol = __get_vma_policy(vma, vma->vm_start);
1659         if (pol) {
1660                 mpol_to_str(buffer, sizeof(buffer), pol);
1661                 mpol_cond_put(pol);
1662         } else {
1663                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1664         }
1665
1666         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1667
1668         if (file) {
1669                 seq_puts(m, " file=");
1670                 seq_file_path(m, file, "\n\t= ");
1671         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1672                 seq_puts(m, " heap");
1673         } else if (is_stack(proc_priv, vma)) {
1674                 seq_puts(m, " stack");
1675         }
1676
1677         if (is_vm_hugetlb_page(vma))
1678                 seq_puts(m, " huge");
1679
1680         /* mmap_sem is held by m_start */
1681         walk_page_vma(vma, &walk);
1682
1683         if (!md->pages)
1684                 goto out;
1685
1686         if (md->anon)
1687                 seq_printf(m, " anon=%lu", md->anon);
1688
1689         if (md->dirty)
1690                 seq_printf(m, " dirty=%lu", md->dirty);
1691
1692         if (md->pages != md->anon && md->pages != md->dirty)
1693                 seq_printf(m, " mapped=%lu", md->pages);
1694
1695         if (md->mapcount_max > 1)
1696                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1697
1698         if (md->swapcache)
1699                 seq_printf(m, " swapcache=%lu", md->swapcache);
1700
1701         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1702                 seq_printf(m, " active=%lu", md->active);
1703
1704         if (md->writeback)
1705                 seq_printf(m, " writeback=%lu", md->writeback);
1706
1707         for_each_node_state(nid, N_MEMORY)
1708                 if (md->node[nid])
1709                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1710
1711         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1712 out:
1713         seq_putc(m, '\n');
1714         m_cache_vma(m, vma);
1715         return 0;
1716 }
1717
1718 static int show_pid_numa_map(struct seq_file *m, void *v)
1719 {
1720         return show_numa_map(m, v, 1);
1721 }
1722
1723 static int show_tid_numa_map(struct seq_file *m, void *v)
1724 {
1725         return show_numa_map(m, v, 0);
1726 }
1727
1728 static const struct seq_operations proc_pid_numa_maps_op = {
1729         .start  = m_start,
1730         .next   = m_next,
1731         .stop   = m_stop,
1732         .show   = show_pid_numa_map,
1733 };
1734
1735 static const struct seq_operations proc_tid_numa_maps_op = {
1736         .start  = m_start,
1737         .next   = m_next,
1738         .stop   = m_stop,
1739         .show   = show_tid_numa_map,
1740 };
1741
1742 static int numa_maps_open(struct inode *inode, struct file *file,
1743                           const struct seq_operations *ops)
1744 {
1745         return proc_maps_open(inode, file, ops,
1746                                 sizeof(struct numa_maps_private));
1747 }
1748
1749 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1750 {
1751         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1752 }
1753
1754 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1755 {
1756         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1757 }
1758
1759 const struct file_operations proc_pid_numa_maps_operations = {
1760         .open           = pid_numa_maps_open,
1761         .read           = seq_read,
1762         .llseek         = seq_lseek,
1763         .release        = proc_map_release,
1764 };
1765
1766 const struct file_operations proc_tid_numa_maps_operations = {
1767         .open           = tid_numa_maps_open,
1768         .read           = seq_read,
1769         .llseek         = seq_lseek,
1770         .release        = proc_map_release,
1771 };
1772 #endif /* CONFIG_NUMA */