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         /* We don't show the stack guard page in /proc/maps */
 302         start = vma->vm_start;
 303         if (stack_guard_page_start(vma, start))
 304                 start += PAGE_SIZE;
 305         end = vma->vm_end;
 306         if (stack_guard_page_end(vma, end))
 307                 end -= PAGE_SIZE;
 308
 309         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 310         seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
 311                         start,
 312                         end,
 313                         flags & VM_READ ? 'r' : '-',
 314                         flags & VM_WRITE ? 'w' : '-',
 315                         flags & VM_EXEC ? 'x' : '-',
 316                         flags & VM_MAYSHARE ? 's' : 'p',
 317                         pgoff,
 318                         MAJOR(dev), MINOR(dev), ino);
 319
 320         /*
 321          * Print the dentry name for named mappings, and a
 322          * special [heap] marker for the heap:
 323          */
 324         if (file) {
 325                 seq_pad(m, ' ');
 326                 seq_file_path(m, file, "\n");
 327                 goto done;
 328         }
 329
 330         if (vma->vm_ops && vma->vm_ops->name) {
 331                 name = vma->vm_ops->name(vma);
 332                 if (name)
 333                         goto done;
 334         }
 335
 336         name = arch_vma_name(vma);
 337         if (!name) {
 338                 if (!mm) {
 339                         name = "[vdso]";
 340                         goto done;
 341                 }
 342
 343                 if (vma->vm_start <= mm->brk &&
 344                     vma->vm_end >= mm->start_brk) {
 345                         name = "[heap]";
 346                         goto done;
 347                 }
 348
 349                 if (is_stack(priv, vma))
 350                         name = "[stack]";
 351         }
 352
 353 done:
 354         if (name) {
 355                 seq_pad(m, ' ');
 356                 seq_puts(m, name);
 357         }
 358         seq_putc(m, '\n');
 359 }
 360
 361 static int show_map(struct seq_file *m, void *v, int is_pid)
 362 {
 363         show_map_vma(m, v, is_pid);
 364         m_cache_vma(m, v);
 365         return 0;
 366 }
 367
 368 static int show_pid_map(struct seq_file *m, void *v)
 369 {
 370         return show_map(m, v, 1);
 371 }
 372
 373 static int show_tid_map(struct seq_file *m, void *v)
 374 {
 375         return show_map(m, v, 0);
 376 }
 377
 378 static const struct seq_operations proc_pid_maps_op = {
 379         .start  = m_start,
 380         .next   = m_next,
 381         .stop   = m_stop,
 382         .show   = show_pid_map
 383 };
 384
 385 static const struct seq_operations proc_tid_maps_op = {
 386         .start  = m_start,
 387         .next   = m_next,
 388         .stop   = m_stop,
 389         .show   = show_tid_map
 390 };
 391
 392 static int pid_maps_open(struct inode *inode, struct file *file)
 393 {
 394         return do_maps_open(inode, file, &proc_pid_maps_op);
 395 }
 396
 397 static int tid_maps_open(struct inode *inode, struct file *file)
 398 {
 399         return do_maps_open(inode, file, &proc_tid_maps_op);
 400 }
 401
 402 const struct file_operations proc_pid_maps_operations = {
 403         .open           = pid_maps_open,
 404         .read           = seq_read,
 405         .llseek         = seq_lseek,
 406         .release        = proc_map_release,
 407 };
 408
 409 const struct file_operations proc_tid_maps_operations = {
 410         .open           = tid_maps_open,
 411         .read           = seq_read,
 412         .llseek         = seq_lseek,
 413         .release        = proc_map_release,
 414 };
 415
 416 /*
 417  * Proportional Set Size(PSS): my share of RSS.
 418  *
 419  * PSS of a process is the count of pages it has in memory, where each
 420  * page is divided by the number of processes sharing it.  So if a
 421  * process has 1000 pages all to itself, and 1000 shared with one other
 422  * process, its PSS will be 1500.
 423  *
 424  * To keep (accumulated) division errors low, we adopt a 64bit
 425  * fixed-point pss counter to minimize division errors. So (pss >>
 426  * PSS_SHIFT) would be the real byte count.
 427  *
 428  * A shift of 12 before division means (assuming 4K page size):
 429  *      - 1M 3-user-pages add up to 8KB errors;
 430  *      - supports mapcount up to 2^24, or 16M;
 431  *      - supports PSS up to 2^52 bytes, or 4PB.
 432  */
 433 #define PSS_SHIFT 12
 434
 435 #ifdef CONFIG_PROC_PAGE_MONITOR
 436 struct mem_size_stats {
 437         unsigned long resident;
 438         unsigned long shared_clean;
 439         unsigned long shared_dirty;
 440         unsigned long private_clean;
 441         unsigned long private_dirty;
 442         unsigned long referenced;
 443         unsigned long anonymous;
 444         unsigned long anonymous_thp;
 445         unsigned long shmem_thp;
 446         unsigned long swap;
 447         unsigned long shared_hugetlb;
 448         unsigned long private_hugetlb;
 449         u64 pss;
 450         u64 swap_pss;
 451         bool check_shmem_swap;
 452 };
 453
 454 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 455                 bool compound, bool young, bool dirty)
 456 {
 457         int i, nr = compound ? 1 << compound_order(page) : 1;
 458         unsigned long size = nr * PAGE_SIZE;
 459
 460         if (PageAnon(page))
 461                 mss->anonymous += size;
 462
 463         mss->resident += size;
 464         /* Accumulate the size in pages that have been accessed. */
 465         if (young || page_is_young(page) || PageReferenced(page))
 466                 mss->referenced += size;
 467
 468         /*
 469          * page_count(page) == 1 guarantees the page is mapped exactly once.
 470          * If any subpage of the compound page mapped with PTE it would elevate
 471          * page_count().
 472          */
 473         if (page_count(page) == 1) {
 474                 if (dirty || PageDirty(page))
 475                         mss->private_dirty += size;
 476                 else
 477                         mss->private_clean += size;
 478                 mss->pss += (u64)size << PSS_SHIFT;
 479                 return;
 480         }
 481
 482         for (i = 0; i < nr; i++, page++) {
 483                 int mapcount = page_mapcount(page);
 484
 485                 if (mapcount >= 2) {
 486                         if (dirty || PageDirty(page))
 487                                 mss->shared_dirty += PAGE_SIZE;
 488                         else
 489                                 mss->shared_clean += PAGE_SIZE;
 490                         mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
 491                 } else {
 492                         if (dirty || PageDirty(page))
 493                                 mss->private_dirty += PAGE_SIZE;
 494                         else
 495                                 mss->private_clean += PAGE_SIZE;
 496                         mss->pss += PAGE_SIZE << PSS_SHIFT;
 497                 }
 498         }
 499 }
 500
 501 #ifdef CONFIG_SHMEM
 502 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 503                 struct mm_walk *walk)
 504 {
 505         struct mem_size_stats *mss = walk->private;
 506
 507         mss->swap += shmem_partial_swap_usage(
 508                         walk->vma->vm_file->f_mapping, addr, end);
 509
 510         return 0;
 511 }
 512 #endif
 513
 514 static void smaps_pte_entry(pte_t *pte, 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 = NULL;
 520
 521         if (pte_present(*pte)) {
 522                 page = vm_normal_page(vma, addr, *pte);
 523         } else if (is_swap_pte(*pte)) {
 524                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 525
 526                 if (!non_swap_entry(swpent)) {
 527                         int mapcount;
 528
 529                         mss->swap += PAGE_SIZE;
 530                         mapcount = swp_swapcount(swpent);
 531                         if (mapcount >= 2) {
 532                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 533
 534                                 do_div(pss_delta, mapcount);
 535                                 mss->swap_pss += pss_delta;
 536                         } else {
 537                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 538                         }
 539                 } else if (is_migration_entry(swpent))
 540                         page = migration_entry_to_page(swpent);
 541         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 542                                                         && pte_none(*pte))) {
 543                 page = find_get_entry(vma->vm_file->f_mapping,
 544                                                 linear_page_index(vma, addr));
 545                 if (!page)
 546                         return;
 547
 548                 if (radix_tree_exceptional_entry(page))
 549                         mss->swap += PAGE_SIZE;
 550                 else
 551                         put_page(page);
 552
 553                 return;
 554         }
 555
 556         if (!page)
 557                 return;
 558
 559         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
 560 }
 561
 562 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 563 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 564                 struct mm_walk *walk)
 565 {
 566         struct mem_size_stats *mss = walk->private;
 567         struct vm_area_struct *vma = walk->vma;
 568         struct page *page;
 569
 570         /* FOLL_DUMP will return -EFAULT on huge zero page */
 571         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 572         if (IS_ERR_OR_NULL(page))
 573                 return;
 574         if (PageAnon(page))
 575                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 576         else if (PageSwapBacked(page))
 577                 mss->shmem_thp += HPAGE_PMD_SIZE;
 578         else if (is_zone_device_page(page))
 579                 /* pass */;
 580         else
 581                 VM_BUG_ON_PAGE(1, page);
 582         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
 583 }
 584 #else
 585 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 586                 struct mm_walk *walk)
 587 {
 588 }
 589 #endif
 590
 591 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 592                            struct mm_walk *walk)
 593 {
 594         struct vm_area_struct *vma = walk->vma;
 595         pte_t *pte;
 596         spinlock_t *ptl;
 597
 598         ptl = pmd_trans_huge_lock(pmd, vma);
 599         if (ptl) {
 600                 smaps_pmd_entry(pmd, addr, walk);
 601                 spin_unlock(ptl);
 602                 return 0;
 603         }
 604
 605         if (pmd_trans_unstable(pmd))
 606                 return 0;
 607         /*
 608          * The mmap_sem held all the way back in m_start() is what
 609          * keeps khugepaged out of here and from collapsing things
 610          * in here.
 611          */
 612         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 613         for (; addr != end; pte++, addr += PAGE_SIZE)
 614                 smaps_pte_entry(pte, addr, walk);
 615         pte_unmap_unlock(pte - 1, ptl);
 616         cond_resched();
 617         return 0;
 618 }
 619
 620 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 621 {
 622         /*
 623          * Don't forget to update Documentation/ on changes.
 624          */
 625         static const char mnemonics[BITS_PER_LONG][2] = {
 626                 /*
 627                  * In case if we meet a flag we don't know about.
 628                  */
 629                 [0 ... (BITS_PER_LONG-1)] = "??",
 630
 631                 [ilog2(VM_READ)]        = "rd",
 632                 [ilog2(VM_WRITE)]       = "wr",
 633                 [ilog2(VM_EXEC)]        = "ex",
 634                 [ilog2(VM_SHARED)]      = "sh",
 635                 [ilog2(VM_MAYREAD)]     = "mr",
 636                 [ilog2(VM_MAYWRITE)]    = "mw",
 637                 [ilog2(VM_MAYEXEC)]     = "me",
 638                 [ilog2(VM_MAYSHARE)]    = "ms",
 639                 [ilog2(VM_GROWSDOWN)]   = "gd",
 640                 [ilog2(VM_PFNMAP)]      = "pf",
 641                 [ilog2(VM_DENYWRITE)]   = "dw",
 642 #ifdef CONFIG_X86_INTEL_MPX
 643                 [ilog2(VM_MPX)]         = "mp",
 644 #endif
 645                 [ilog2(VM_LOCKED)]      = "lo",
 646                 [ilog2(VM_IO)]          = "io",
 647                 [ilog2(VM_SEQ_READ)]    = "sr",
 648                 [ilog2(VM_RAND_READ)]   = "rr",
 649                 [ilog2(VM_DONTCOPY)]    = "dc",
 650                 [ilog2(VM_DONTEXPAND)]  = "de",
 651                 [ilog2(VM_ACCOUNT)]     = "ac",
 652                 [ilog2(VM_NORESERVE)]   = "nr",
 653                 [ilog2(VM_HUGETLB)]     = "ht",
 654                 [ilog2(VM_ARCH_1)]      = "ar",
 655                 [ilog2(VM_DONTDUMP)]    = "dd",
 656 #ifdef CONFIG_MEM_SOFT_DIRTY
 657                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 658 #endif
 659                 [ilog2(VM_MIXEDMAP)]    = "mm",
 660                 [ilog2(VM_HUGEPAGE)]    = "hg",
 661                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 662                 [ilog2(VM_MERGEABLE)]   = "mg",
 663                 [ilog2(VM_UFFD_MISSING)]= "um",
 664                 [ilog2(VM_UFFD_WP)]     = "uw",
 665 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 666                 /* These come out via ProtectionKey: */
 667                 [ilog2(VM_PKEY_BIT0)]   = "",
 668                 [ilog2(VM_PKEY_BIT1)]   = "",
 669                 [ilog2(VM_PKEY_BIT2)]   = "",
 670                 [ilog2(VM_PKEY_BIT3)]   = "",
 671 #endif
 672         };
 673         size_t i;
 674
 675         seq_puts(m, "VmFlags: ");
 676         for (i = 0; i < BITS_PER_LONG; i++) {
 677                 if (!mnemonics[i][0])
 678                         continue;
 679                 if (vma->vm_flags & (1UL << i)) {
 680                         seq_printf(m, "%c%c ",
 681                                    mnemonics[i][0], mnemonics[i][1]);
 682                 }
 683         }
 684         seq_putc(m, '\n');
 685 }
 686
 687 #ifdef CONFIG_HUGETLB_PAGE
 688 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 689                                  unsigned long addr, unsigned long end,
 690                                  struct mm_walk *walk)
 691 {
 692         struct mem_size_stats *mss = walk->private;
 693         struct vm_area_struct *vma = walk->vma;
 694         struct page *page = NULL;
 695
 696         if (pte_present(*pte)) {
 697                 page = vm_normal_page(vma, addr, *pte);
 698         } else if (is_swap_pte(*pte)) {
 699                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 700
 701                 if (is_migration_entry(swpent))
 702                         page = migration_entry_to_page(swpent);
 703         }
 704         if (page) {
 705                 int mapcount = page_mapcount(page);
 706
 707                 if (mapcount >= 2)
 708                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 709                 else
 710                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 711         }
 712         return 0;
 713 }
 714 #endif /* HUGETLB_PAGE */
 715
 716 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
 717 {
 718 }
 719
 720 static int show_smap(struct seq_file *m, void *v, int is_pid)
 721 {
 722         struct vm_area_struct *vma = v;
 723         struct mem_size_stats mss;
 724         struct mm_walk smaps_walk = {
 725                 .pmd_entry = smaps_pte_range,
 726 #ifdef CONFIG_HUGETLB_PAGE
 727                 .hugetlb_entry = smaps_hugetlb_range,
 728 #endif
 729                 .mm = vma->vm_mm,
 730                 .private = &mss,
 731         };
 732
 733         memset(&mss, 0, sizeof mss);
 734
 735 #ifdef CONFIG_SHMEM
 736         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 737                 /*
 738                  * For shared or readonly shmem mappings we know that all
 739                  * swapped out pages belong to the shmem object, and we can
 740                  * obtain the swap value much more efficiently. For private
 741                  * writable mappings, we might have COW pages that are
 742                  * not affected by the parent swapped out pages of the shmem
 743                  * object, so we have to distinguish them during the page walk.
 744                  * Unless we know that the shmem object (or the part mapped by
 745                  * our VMA) has no swapped out pages at all.
 746                  */
 747                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 748
 749                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 750                                         !(vma->vm_flags & VM_WRITE)) {
 751                         mss.swap = shmem_swapped;
 752                 } else {
 753                         mss.check_shmem_swap = true;
 754                         smaps_walk.pte_hole = smaps_pte_hole;
 755                 }
 756         }
 757 #endif
 758
 759         /* mmap_sem is held in m_start */
 760         walk_page_vma(vma, &smaps_walk);
 761
 762         show_map_vma(m, vma, is_pid);
 763
 764         seq_printf(m,
 765                    "Size:           %8lu kB\n"
 766                    "Rss:            %8lu kB\n"
 767                    "Pss:            %8lu kB\n"
 768                    "Shared_Clean:   %8lu kB\n"
 769                    "Shared_Dirty:   %8lu kB\n"
 770                    "Private_Clean:  %8lu kB\n"
 771                    "Private_Dirty:  %8lu kB\n"
 772                    "Referenced:     %8lu kB\n"
 773                    "Anonymous:      %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.anonymous_thp >> 10,
 793                    mss.shmem_thp >> 10,
 794                    mss.shared_hugetlb >> 10,
 795                    mss.private_hugetlb >> 10,
 796                    mss.swap >> 10,
 797                    (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
 798                    vma_kernel_pagesize(vma) >> 10,
 799                    vma_mmu_pagesize(vma) >> 10,
 800                    (vma->vm_flags & VM_LOCKED) ?
 801                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 802
 803         arch_show_smap(m, vma);
 804         show_smap_vma_flags(m, vma);
 805         m_cache_vma(m, vma);
 806         return 0;
 807 }
 808
 809 static int show_pid_smap(struct seq_file *m, void *v)
 810 {
 811         return show_smap(m, v, 1);
 812 }
 813
 814 static int show_tid_smap(struct seq_file *m, void *v)
 815 {
 816         return show_smap(m, v, 0);
 817 }
 818
 819 static const struct seq_operations proc_pid_smaps_op = {
 820         .start  = m_start,
 821         .next   = m_next,
 822         .stop   = m_stop,
 823         .show   = show_pid_smap
 824 };
 825
 826 static const struct seq_operations proc_tid_smaps_op = {
 827         .start  = m_start,
 828         .next   = m_next,
 829         .stop   = m_stop,
 830         .show   = show_tid_smap
 831 };
 832
 833 static int pid_smaps_open(struct inode *inode, struct file *file)
 834 {
 835         return do_maps_open(inode, file, &proc_pid_smaps_op);
 836 }
 837
 838 static int tid_smaps_open(struct inode *inode, struct file *file)
 839 {
 840         return do_maps_open(inode, file, &proc_tid_smaps_op);
 841 }
 842
 843 const struct file_operations proc_pid_smaps_operations = {
 844         .open           = pid_smaps_open,
 845         .read           = seq_read,
 846         .llseek         = seq_lseek,
 847         .release        = proc_map_release,
 848 };
 849
 850 const struct file_operations proc_tid_smaps_operations = {
 851         .open           = tid_smaps_open,
 852         .read           = seq_read,
 853         .llseek         = seq_lseek,
 854         .release        = proc_map_release,
 855 };
 856
 857 enum clear_refs_types {
 858         CLEAR_REFS_ALL = 1,
 859         CLEAR_REFS_ANON,
 860         CLEAR_REFS_MAPPED,
 861         CLEAR_REFS_SOFT_DIRTY,
 862         CLEAR_REFS_MM_HIWATER_RSS,
 863         CLEAR_REFS_LAST,
 864 };
 865
 866 struct clear_refs_private {
 867         enum clear_refs_types type;
 868 };
 869
 870 #ifdef CONFIG_MEM_SOFT_DIRTY
 871 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 872                 unsigned long addr, pte_t *pte)
 873 {
 874         /*
 875          * The soft-dirty tracker uses #PF-s to catch writes
 876          * to pages, so write-protect the pte as well. See the
 877          * Documentation/vm/soft-dirty.txt for full description
 878          * of how soft-dirty works.
 879          */
 880         pte_t ptent = *pte;
 881
 882         if (pte_present(ptent)) {
 883                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 884                 ptent = pte_wrprotect(ptent);
 885                 ptent = pte_clear_soft_dirty(ptent);
 886                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 887         } else if (is_swap_pte(ptent)) {
 888                 ptent = pte_swp_clear_soft_dirty(ptent);
 889                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 890         }
 891 }
 892 #else
 893 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 894                 unsigned long addr, pte_t *pte)
 895 {
 896 }
 897 #endif
 898
 899 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 900 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 901                 unsigned long addr, pmd_t *pmdp)
 902 {
 903         pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
 904
 905         pmd = pmd_wrprotect(pmd);
 906         pmd = pmd_clear_soft_dirty(pmd);
 907
 908         set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 909 }
 910 #else
 911 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 912                 unsigned long addr, pmd_t *pmdp)
 913 {
 914 }
 915 #endif
 916
 917 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 918                                 unsigned long end, struct mm_walk *walk)
 919 {
 920         struct clear_refs_private *cp = walk->private;
 921         struct vm_area_struct *vma = walk->vma;
 922         pte_t *pte, ptent;
 923         spinlock_t *ptl;
 924         struct page *page;
 925
 926         ptl = pmd_trans_huge_lock(pmd, vma);
 927         if (ptl) {
 928                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 929                         clear_soft_dirty_pmd(vma, addr, pmd);
 930                         goto out;
 931                 }
 932
 933                 page = pmd_page(*pmd);
 934
 935                 /* Clear accessed and referenced bits. */
 936                 pmdp_test_and_clear_young(vma, addr, pmd);
 937                 test_and_clear_page_young(page);
 938                 ClearPageReferenced(page);
 939 out:
 940                 spin_unlock(ptl);
 941                 return 0;
 942         }
 943
 944         if (pmd_trans_unstable(pmd))
 945                 return 0;
 946
 947         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 948         for (; addr != end; pte++, addr += PAGE_SIZE) {
 949                 ptent = *pte;
 950
 951                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 952                         clear_soft_dirty(vma, addr, pte);
 953                         continue;
 954                 }
 955
 956                 if (!pte_present(ptent))
 957                         continue;
 958
 959                 page = vm_normal_page(vma, addr, ptent);
 960                 if (!page)
 961                         continue;
 962
 963                 /* Clear accessed and referenced bits. */
 964                 ptep_test_and_clear_young(vma, addr, pte);
 965                 test_and_clear_page_young(page);
 966                 ClearPageReferenced(page);
 967         }
 968         pte_unmap_unlock(pte - 1, ptl);
 969         cond_resched();
 970         return 0;
 971 }
 972
 973 static int clear_refs_test_walk(unsigned long start, unsigned long end,
 974                                 struct mm_walk *walk)
 975 {
 976         struct clear_refs_private *cp = walk->private;
 977         struct vm_area_struct *vma = walk->vma;
 978
 979         if (vma->vm_flags & VM_PFNMAP)
 980                 return 1;
 981
 982         /*
 983          * Writing 1 to /proc/pid/clear_refs affects all pages.
 984          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
 985          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
 986          * Writing 4 to /proc/pid/clear_refs affects all pages.
 987          */
 988         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
 989                 return 1;
 990         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
 991                 return 1;
 992         return 0;
 993 }
 994
 995 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 996                                 size_t count, loff_t *ppos)
 997 {
 998         struct task_struct *task;
 999         char buffer[PROC_NUMBUF];
1000         struct mm_struct *mm;
1001         struct vm_area_struct *vma;
1002         enum clear_refs_types type;
1003         int itype;
1004         int rv;
1005
1006         memset(buffer, 0, sizeof(buffer));
1007         if (count > sizeof(buffer) - 1)
1008                 count = sizeof(buffer) - 1;
1009         if (copy_from_user(buffer, buf, count))
1010                 return -EFAULT;
1011         rv = kstrtoint(strstrip(buffer), 10, &itype);
1012         if (rv < 0)
1013                 return rv;
1014         type = (enum clear_refs_types)itype;
1015         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1016                 return -EINVAL;
1017
1018         task = get_proc_task(file_inode(file));
1019         if (!task)
1020                 return -ESRCH;
1021         mm = get_task_mm(task);
1022         if (mm) {
1023                 struct clear_refs_private cp = {
1024                         .type = type,
1025                 };
1026                 struct mm_walk clear_refs_walk = {
1027                         .pmd_entry = clear_refs_pte_range,
1028                         .test_walk = clear_refs_test_walk,
1029                         .mm = mm,
1030                         .private = &cp,
1031                 };
1032
1033                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1034                         if (down_write_killable(&mm->mmap_sem)) {
1035                                 count = -EINTR;
1036                                 goto out_mm;
1037                         }
1038
1039                         /*
1040                          * Writing 5 to /proc/pid/clear_refs resets the peak
1041                          * resident set size to this mm's current rss value.
1042                          */
1043                         reset_mm_hiwater_rss(mm);
1044                         up_write(&mm->mmap_sem);
1045                         goto out_mm;
1046                 }
1047
1048                 down_read(&mm->mmap_sem);
1049                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1050                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1051                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1052                                         continue;
1053                                 up_read(&mm->mmap_sem);
1054                                 if (down_write_killable(&mm->mmap_sem)) {
1055                                         count = -EINTR;
1056                                         goto out_mm;
1057                                 }
1058                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1059                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1060                                         vma_set_page_prot(vma);
1061                                 }
1062                                 downgrade_write(&mm->mmap_sem);
1063                                 break;
1064                         }
1065                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1066                 }
1067                 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1068                 if (type == CLEAR_REFS_SOFT_DIRTY)
1069                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1070                 flush_tlb_mm(mm);
1071                 up_read(&mm->mmap_sem);
1072 out_mm:
1073                 mmput(mm);
1074         }
1075         put_task_struct(task);
1076
1077         return count;
1078 }
1079
1080 const struct file_operations proc_clear_refs_operations = {
1081         .write          = clear_refs_write,
1082         .llseek         = noop_llseek,
1083 };
1084
1085 typedef struct {
1086         u64 pme;
1087 } pagemap_entry_t;
1088
1089 struct pagemapread {
1090         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1091         pagemap_entry_t *buffer;
1092         bool show_pfn;
1093 };
1094
1095 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1096 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1097
1098 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1099 #define PM_PFRAME_BITS          55
1100 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1101 #define PM_SOFT_DIRTY           BIT_ULL(55)
1102 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1103 #define PM_FILE                 BIT_ULL(61)
1104 #define PM_SWAP                 BIT_ULL(62)
1105 #define PM_PRESENT              BIT_ULL(63)
1106
1107 #define PM_END_OF_BUFFER    1
1108
1109 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1110 {
1111         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1112 }
1113
1114 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1115                           struct pagemapread *pm)
1116 {
1117         pm->buffer[pm->pos++] = *pme;
1118         if (pm->pos >= pm->len)
1119                 return PM_END_OF_BUFFER;
1120         return 0;
1121 }
1122
1123 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1124                                 struct mm_walk *walk)
1125 {
1126         struct pagemapread *pm = walk->private;
1127         unsigned long addr = start;
1128         int err = 0;
1129
1130         while (addr < end) {
1131                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1132                 pagemap_entry_t pme = make_pme(0, 0);
1133                 /* End of address space hole, which we mark as non-present. */
1134                 unsigned long hole_end;
1135
1136                 if (vma)
1137                         hole_end = min(end, vma->vm_start);
1138                 else
1139                         hole_end = end;
1140
1141                 for (; addr < hole_end; addr += PAGE_SIZE) {
1142                         err = add_to_pagemap(addr, &pme, pm);
1143                         if (err)
1144                                 goto out;
1145                 }
1146
1147                 if (!vma)
1148                         break;
1149
1150                 /* Addresses in the VMA. */
1151                 if (vma->vm_flags & VM_SOFTDIRTY)
1152                         pme = make_pme(0, PM_SOFT_DIRTY);
1153                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1154                         err = add_to_pagemap(addr, &pme, pm);
1155                         if (err)
1156                                 goto out;
1157                 }
1158         }
1159 out:
1160         return err;
1161 }
1162
1163 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1164                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1165 {
1166         u64 frame = 0, flags = 0;
1167         struct page *page = NULL;
1168
1169         if (pte_present(pte)) {
1170                 if (pm->show_pfn)
1171                         frame = pte_pfn(pte);
1172                 flags |= PM_PRESENT;
1173                 page = vm_normal_page(vma, addr, pte);
1174                 if (pte_soft_dirty(pte))
1175                         flags |= PM_SOFT_DIRTY;
1176         } else if (is_swap_pte(pte)) {
1177                 swp_entry_t entry;
1178                 if (pte_swp_soft_dirty(pte))
1179                         flags |= PM_SOFT_DIRTY;
1180                 entry = pte_to_swp_entry(pte);
1181                 frame = swp_type(entry) |
1182                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1183                 flags |= PM_SWAP;
1184                 if (is_migration_entry(entry))
1185                         page = migration_entry_to_page(entry);
1186         }
1187
1188         if (page && !PageAnon(page))
1189                 flags |= PM_FILE;
1190         if (page && page_mapcount(page) == 1)
1191                 flags |= PM_MMAP_EXCLUSIVE;
1192         if (vma->vm_flags & VM_SOFTDIRTY)
1193                 flags |= PM_SOFT_DIRTY;
1194
1195         return make_pme(frame, flags);
1196 }
1197
1198 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1199                              struct mm_walk *walk)
1200 {
1201         struct vm_area_struct *vma = walk->vma;
1202         struct pagemapread *pm = walk->private;
1203         spinlock_t *ptl;
1204         pte_t *pte, *orig_pte;
1205         int err = 0;
1206
1207 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1208         ptl = pmd_trans_huge_lock(pmdp, vma);
1209         if (ptl) {
1210                 u64 flags = 0, frame = 0;
1211                 pmd_t pmd = *pmdp;
1212
1213                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1214                         flags |= PM_SOFT_DIRTY;
1215
1216                 /*
1217                  * Currently pmd for thp is always present because thp
1218                  * can not be swapped-out, migrated, or HWPOISONed
1219                  * (split in such cases instead.)
1220                  * This if-check is just to prepare for future implementation.
1221                  */
1222                 if (pmd_present(pmd)) {
1223                         struct page *page = pmd_page(pmd);
1224
1225                         if (page_mapcount(page) == 1)
1226                                 flags |= PM_MMAP_EXCLUSIVE;
1227
1228                         flags |= PM_PRESENT;
1229                         if (pm->show_pfn)
1230                                 frame = pmd_pfn(pmd) +
1231                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1232                 }
1233
1234                 for (; addr != end; addr += PAGE_SIZE) {
1235                         pagemap_entry_t pme = make_pme(frame, flags);
1236
1237                         err = add_to_pagemap(addr, &pme, pm);
1238                         if (err)
1239                                 break;
1240                         if (pm->show_pfn && (flags & PM_PRESENT))
1241                                 frame++;
1242                 }
1243                 spin_unlock(ptl);
1244                 return err;
1245         }
1246
1247         if (pmd_trans_unstable(pmdp))
1248                 return 0;
1249 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1250
1251         /*
1252          * We can assume that @vma always points to a valid one and @end never
1253          * goes beyond vma->vm_end.
1254          */
1255         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1256         for (; addr < end; pte++, addr += PAGE_SIZE) {
1257                 pagemap_entry_t pme;
1258
1259                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1260                 err = add_to_pagemap(addr, &pme, pm);
1261                 if (err)
1262                         break;
1263         }
1264         pte_unmap_unlock(orig_pte, ptl);
1265
1266         cond_resched();
1267
1268         return err;
1269 }
1270
1271 #ifdef CONFIG_HUGETLB_PAGE
1272 /* This function walks within one hugetlb entry in the single call */
1273 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1274                                  unsigned long addr, unsigned long end,
1275                                  struct mm_walk *walk)
1276 {
1277         struct pagemapread *pm = walk->private;
1278         struct vm_area_struct *vma = walk->vma;
1279         u64 flags = 0, frame = 0;
1280         int err = 0;
1281         pte_t pte;
1282
1283         if (vma->vm_flags & VM_SOFTDIRTY)
1284                 flags |= PM_SOFT_DIRTY;
1285
1286         pte = huge_ptep_get(ptep);
1287         if (pte_present(pte)) {
1288                 struct page *page = pte_page(pte);
1289
1290                 if (!PageAnon(page))
1291                         flags |= PM_FILE;
1292
1293                 if (page_mapcount(page) == 1)
1294                         flags |= PM_MMAP_EXCLUSIVE;
1295
1296                 flags |= PM_PRESENT;
1297                 if (pm->show_pfn)
1298                         frame = pte_pfn(pte) +
1299                                 ((addr & ~hmask) >> PAGE_SHIFT);
1300         }
1301
1302         for (; addr != end; addr += PAGE_SIZE) {
1303                 pagemap_entry_t pme = make_pme(frame, flags);
1304
1305                 err = add_to_pagemap(addr, &pme, pm);
1306                 if (err)
1307                         return err;
1308                 if (pm->show_pfn && (flags & PM_PRESENT))
1309                         frame++;
1310         }
1311
1312         cond_resched();
1313
1314         return err;
1315 }
1316 #endif /* HUGETLB_PAGE */
1317
1318 /*
1319  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1320  *
1321  * For each page in the address space, this file contains one 64-bit entry
1322  * consisting of the following:
1323  *
1324  * Bits 0-54  page frame number (PFN) if present
1325  * Bits 0-4   swap type if swapped
1326  * Bits 5-54  swap offset if swapped
1327  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1328  * Bit  56    page exclusively mapped
1329  * Bits 57-60 zero
1330  * Bit  61    page is file-page or shared-anon
1331  * Bit  62    page swapped
1332  * Bit  63    page present
1333  *
1334  * If the page is not present but in swap, then the PFN contains an
1335  * encoding of the swap file number and the page's offset into the
1336  * swap. Unmapped pages return a null PFN. This allows determining
1337  * precisely which pages are mapped (or in swap) and comparing mapped
1338  * pages between processes.
1339  *
1340  * Efficient users of this interface will use /proc/pid/maps to
1341  * determine which areas of memory are actually mapped and llseek to
1342  * skip over unmapped regions.
1343  */
1344 static ssize_t pagemap_read(struct file *file, char __user *buf,
1345                             size_t count, loff_t *ppos)
1346 {
1347         struct mm_struct *mm = file->private_data;
1348         struct pagemapread pm;
1349         struct mm_walk pagemap_walk = {};
1350         unsigned long src;
1351         unsigned long svpfn;
1352         unsigned long start_vaddr;
1353         unsigned long end_vaddr;
1354         int ret = 0, copied = 0;
1355
1356         if (!mm || !mmget_not_zero(mm))
1357                 goto out;
1358
1359         ret = -EINVAL;
1360         /* file position must be aligned */
1361         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1362                 goto out_mm;
1363
1364         ret = 0;
1365         if (!count)
1366                 goto out_mm;
1367
1368         /* do not disclose physical addresses: attack vector */
1369         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1370
1371         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1372         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1373         ret = -ENOMEM;
1374         if (!pm.buffer)
1375                 goto out_mm;
1376
1377         pagemap_walk.pmd_entry = pagemap_pmd_range;
1378         pagemap_walk.pte_hole = pagemap_pte_hole;
1379 #ifdef CONFIG_HUGETLB_PAGE
1380         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1381 #endif
1382         pagemap_walk.mm = mm;
1383         pagemap_walk.private = &pm;
1384
1385         src = *ppos;
1386         svpfn = src / PM_ENTRY_BYTES;
1387         start_vaddr = svpfn << PAGE_SHIFT;
1388         end_vaddr = mm->task_size;
1389
1390         /* watch out for wraparound */
1391         if (svpfn > mm->task_size >> PAGE_SHIFT)
1392                 start_vaddr = end_vaddr;
1393
1394         /*
1395          * The odds are that this will stop walking way
1396          * before end_vaddr, because the length of the
1397          * user buffer is tracked in "pm", and the walk
1398          * will stop when we hit the end of the buffer.
1399          */
1400         ret = 0;
1401         while (count && (start_vaddr < end_vaddr)) {
1402                 int len;
1403                 unsigned long end;
1404
1405                 pm.pos = 0;
1406                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1407                 /* overflow ? */
1408                 if (end < start_vaddr || end > end_vaddr)
1409                         end = end_vaddr;
1410                 down_read(&mm->mmap_sem);
1411                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1412                 up_read(&mm->mmap_sem);
1413                 start_vaddr = end;
1414
1415                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1416                 if (copy_to_user(buf, pm.buffer, len)) {
1417                         ret = -EFAULT;
1418                         goto out_free;
1419                 }
1420                 copied += len;
1421                 buf += len;
1422                 count -= len;
1423         }
1424         *ppos += copied;
1425         if (!ret || ret == PM_END_OF_BUFFER)
1426                 ret = copied;
1427
1428 out_free:
1429         kfree(pm.buffer);
1430 out_mm:
1431         mmput(mm);
1432 out:
1433         return ret;
1434 }
1435
1436 static int pagemap_open(struct inode *inode, struct file *file)
1437 {
1438         struct mm_struct *mm;
1439
1440         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1441         if (IS_ERR(mm))
1442                 return PTR_ERR(mm);
1443         file->private_data = mm;
1444         return 0;
1445 }
1446
1447 static int pagemap_release(struct inode *inode, struct file *file)
1448 {
1449         struct mm_struct *mm = file->private_data;
1450
1451         if (mm)
1452                 mmdrop(mm);
1453         return 0;
1454 }
1455
1456 const struct file_operations proc_pagemap_operations = {
1457         .llseek         = mem_lseek, /* borrow this */
1458         .read           = pagemap_read,
1459         .open           = pagemap_open,
1460         .release        = pagemap_release,
1461 };
1462 #endif /* CONFIG_PROC_PAGE_MONITOR */
1463
1464 #ifdef CONFIG_NUMA
1465
1466 struct numa_maps {
1467         unsigned long pages;
1468         unsigned long anon;
1469         unsigned long active;
1470         unsigned long writeback;
1471         unsigned long mapcount_max;
1472         unsigned long dirty;
1473         unsigned long swapcache;
1474         unsigned long node[MAX_NUMNODES];
1475 };
1476
1477 struct numa_maps_private {
1478         struct proc_maps_private proc_maps;
1479         struct numa_maps md;
1480 };
1481
1482 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1483                         unsigned long nr_pages)
1484 {
1485         int count = page_mapcount(page);
1486
1487         md->pages += nr_pages;
1488         if (pte_dirty || PageDirty(page))
1489                 md->dirty += nr_pages;
1490
1491         if (PageSwapCache(page))
1492                 md->swapcache += nr_pages;
1493
1494         if (PageActive(page) || PageUnevictable(page))
1495                 md->active += nr_pages;
1496
1497         if (PageWriteback(page))
1498                 md->writeback += nr_pages;
1499
1500         if (PageAnon(page))
1501                 md->anon += nr_pages;
1502
1503         if (count > md->mapcount_max)
1504                 md->mapcount_max = count;
1505
1506         md->node[page_to_nid(page)] += nr_pages;
1507 }
1508
1509 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1510                 unsigned long addr)
1511 {
1512         struct page *page;
1513         int nid;
1514
1515         if (!pte_present(pte))
1516                 return NULL;
1517
1518         page = vm_normal_page(vma, addr, pte);
1519         if (!page)
1520                 return NULL;
1521
1522         if (PageReserved(page))
1523                 return NULL;
1524
1525         nid = page_to_nid(page);
1526         if (!node_isset(nid, node_states[N_MEMORY]))
1527                 return NULL;
1528
1529         return page;
1530 }
1531
1532 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1533 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1534                                               struct vm_area_struct *vma,
1535                                               unsigned long addr)
1536 {
1537         struct page *page;
1538         int nid;
1539
1540         if (!pmd_present(pmd))
1541                 return NULL;
1542
1543         page = vm_normal_page_pmd(vma, addr, pmd);
1544         if (!page)
1545                 return NULL;
1546
1547         if (PageReserved(page))
1548                 return NULL;
1549
1550         nid = page_to_nid(page);
1551         if (!node_isset(nid, node_states[N_MEMORY]))
1552                 return NULL;
1553
1554         return page;
1555 }
1556 #endif
1557
1558 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1559                 unsigned long end, struct mm_walk *walk)
1560 {
1561         struct numa_maps *md = walk->private;
1562         struct vm_area_struct *vma = walk->vma;
1563         spinlock_t *ptl;
1564         pte_t *orig_pte;
1565         pte_t *pte;
1566
1567 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1568         ptl = pmd_trans_huge_lock(pmd, vma);
1569         if (ptl) {
1570                 struct page *page;
1571
1572                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1573                 if (page)
1574                         gather_stats(page, md, pmd_dirty(*pmd),
1575                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1576                 spin_unlock(ptl);
1577                 return 0;
1578         }
1579
1580         if (pmd_trans_unstable(pmd))
1581                 return 0;
1582 #endif
1583         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1584         do {
1585                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1586                 if (!page)
1587                         continue;
1588                 gather_stats(page, md, pte_dirty(*pte), 1);
1589
1590         } while (pte++, addr += PAGE_SIZE, addr != end);
1591         pte_unmap_unlock(orig_pte, ptl);
1592         cond_resched();
1593         return 0;
1594 }
1595 #ifdef CONFIG_HUGETLB_PAGE
1596 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1597                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1598 {
1599         pte_t huge_pte = huge_ptep_get(pte);
1600         struct numa_maps *md;
1601         struct page *page;
1602
1603         if (!pte_present(huge_pte))
1604                 return 0;
1605
1606         page = pte_page(huge_pte);
1607         if (!page)
1608                 return 0;
1609
1610         md = walk->private;
1611         gather_stats(page, md, pte_dirty(huge_pte), 1);
1612         return 0;
1613 }
1614
1615 #else
1616 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1617                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1618 {
1619         return 0;
1620 }
1621 #endif
1622
1623 /*
1624  * Display pages allocated per node and memory policy via /proc.
1625  */
1626 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1627 {
1628         struct numa_maps_private *numa_priv = m->private;
1629         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1630         struct vm_area_struct *vma = v;
1631         struct numa_maps *md = &numa_priv->md;
1632         struct file *file = vma->vm_file;
1633         struct mm_struct *mm = vma->vm_mm;
1634         struct mm_walk walk = {
1635                 .hugetlb_entry = gather_hugetlb_stats,
1636                 .pmd_entry = gather_pte_stats,
1637                 .private = md,
1638                 .mm = mm,
1639         };
1640         struct mempolicy *pol;
1641         char buffer[64];
1642         int nid;
1643
1644         if (!mm)
1645                 return 0;
1646
1647         /* Ensure we start with an empty set of numa_maps statistics. */
1648         memset(md, 0, sizeof(*md));
1649
1650         pol = __get_vma_policy(vma, vma->vm_start);
1651         if (pol) {
1652                 mpol_to_str(buffer, sizeof(buffer), pol);
1653                 mpol_cond_put(pol);
1654         } else {
1655                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1656         }
1657
1658         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1659
1660         if (file) {
1661                 seq_puts(m, " file=");
1662                 seq_file_path(m, file, "\n\t= ");
1663         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1664                 seq_puts(m, " heap");
1665         } else if (is_stack(proc_priv, vma)) {
1666                 seq_puts(m, " stack");
1667         }
1668
1669         if (is_vm_hugetlb_page(vma))
1670                 seq_puts(m, " huge");
1671
1672         /* mmap_sem is held by m_start */
1673         walk_page_vma(vma, &walk);
1674
1675         if (!md->pages)
1676                 goto out;
1677
1678         if (md->anon)
1679                 seq_printf(m, " anon=%lu", md->anon);
1680
1681         if (md->dirty)
1682                 seq_printf(m, " dirty=%lu", md->dirty);
1683
1684         if (md->pages != md->anon && md->pages != md->dirty)
1685                 seq_printf(m, " mapped=%lu", md->pages);
1686
1687         if (md->mapcount_max > 1)
1688                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1689
1690         if (md->swapcache)
1691                 seq_printf(m, " swapcache=%lu", md->swapcache);
1692
1693         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1694                 seq_printf(m, " active=%lu", md->active);
1695
1696         if (md->writeback)
1697                 seq_printf(m, " writeback=%lu", md->writeback);
1698
1699         for_each_node_state(nid, N_MEMORY)
1700                 if (md->node[nid])
1701                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1702
1703         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1704 out:
1705         seq_putc(m, '\n');
1706         m_cache_vma(m, vma);
1707         return 0;
1708 }
1709
1710 static int show_pid_numa_map(struct seq_file *m, void *v)
1711 {
1712         return show_numa_map(m, v, 1);
1713 }
1714
1715 static int show_tid_numa_map(struct seq_file *m, void *v)
1716 {
1717         return show_numa_map(m, v, 0);
1718 }
1719
1720 static const struct seq_operations proc_pid_numa_maps_op = {
1721         .start  = m_start,
1722         .next   = m_next,
1723         .stop   = m_stop,
1724         .show   = show_pid_numa_map,
1725 };
1726
1727 static const struct seq_operations proc_tid_numa_maps_op = {
1728         .start  = m_start,
1729         .next   = m_next,
1730         .stop   = m_stop,
1731         .show   = show_tid_numa_map,
1732 };
1733
1734 static int numa_maps_open(struct inode *inode, struct file *file,
1735                           const struct seq_operations *ops)
1736 {
1737         return proc_maps_open(inode, file, ops,
1738                                 sizeof(struct numa_maps_private));
1739 }
1740
1741 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1742 {
1743         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1744 }
1745
1746 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1747 {
1748         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1749 }
1750
1751 const struct file_operations proc_pid_numa_maps_operations = {
1752         .open           = pid_numa_maps_open,
1753         .read           = seq_read,
1754         .llseek         = seq_lseek,
1755         .release        = proc_map_release,
1756 };
1757
1758 const struct file_operations proc_tid_numa_maps_operations = {
1759         .open           = tid_numa_maps_open,
1760         .read           = seq_read,
1761         .llseek         = seq_lseek,
1762         .release        = proc_map_release,
1763 };
1764 #endif /* CONFIG_NUMA */