fs/proc/task_mmu.c

   1 #include <linux/mm.h>
   2 #include <linux/vmacache.h>
   3 #include <linux/hugetlb.h>
   4 #include <linux/huge_mm.h>
   5 #include <linux/mount.h>
   6 #include <linux/seq_file.h>
   7 #include <linux/highmem.h>
   8 #include <linux/ptrace.h>
   9 #include <linux/slab.h>
  10 #include <linux/pagemap.h>
  11 #include <linux/mempolicy.h>
  12 #include <linux/rmap.h>
  13 #include <linux/swap.h>
  14 #include <linux/swapops.h>
  15 #include <linux/mmu_notifier.h>
  16 #include <linux/page_idle.h>
  17 #include <linux/shmem_fs.h>
  18
  19 #include <asm/elf.h>
  20 #include <asm/uaccess.h>
  21 #include <asm/tlbflush.h>
  22 #include "internal.h"
  23
  24 void task_mem(struct seq_file *m, struct mm_struct *mm)
  25 {
  26         unsigned long text, lib, swap, ptes, pmds, anon, file, shmem;
  27         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  28
  29         anon = get_mm_counter(mm, MM_ANONPAGES);
  30         file = get_mm_counter(mm, MM_FILEPAGES);
  31         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  32
  33         /*
  34          * Note: to minimize their overhead, mm maintains hiwater_vm and
  35          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  36          * collector of these hiwater stats must therefore get total_vm
  37          * and rss too, which will usually be the higher.  Barriers? not
  38          * worth the effort, such snapshots can always be inconsistent.
  39          */
  40         hiwater_vm = total_vm = mm->total_vm;
  41         if (hiwater_vm < mm->hiwater_vm)
  42                 hiwater_vm = mm->hiwater_vm;
  43         hiwater_rss = total_rss = anon + file + shmem;
  44         if (hiwater_rss < mm->hiwater_rss)
  45                 hiwater_rss = mm->hiwater_rss;
  46
  47         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  48         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  49         swap = get_mm_counter(mm, MM_SWAPENTS);
  50         ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
  51         pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
  52         seq_printf(m,
  53                 "VmPeak:\t%8lu kB\n"
  54                 "VmSize:\t%8lu kB\n"
  55                 "VmLck:\t%8lu kB\n"
  56                 "VmPin:\t%8lu kB\n"
  57                 "VmHWM:\t%8lu kB\n"
  58                 "VmRSS:\t%8lu kB\n"
  59                 "RssAnon:\t%8lu kB\n"
  60                 "RssFile:\t%8lu kB\n"
  61                 "RssShmem:\t%8lu kB\n"
  62                 "VmData:\t%8lu kB\n"
  63                 "VmStk:\t%8lu kB\n"
  64                 "VmExe:\t%8lu kB\n"
  65                 "VmLib:\t%8lu kB\n"
  66                 "VmPTE:\t%8lu kB\n"
  67                 "VmPMD:\t%8lu kB\n"
  68                 "VmSwap:\t%8lu kB\n",
  69                 hiwater_vm << (PAGE_SHIFT-10),
  70                 total_vm << (PAGE_SHIFT-10),
  71                 mm->locked_vm << (PAGE_SHIFT-10),
  72                 mm->pinned_vm << (PAGE_SHIFT-10),
  73                 hiwater_rss << (PAGE_SHIFT-10),
  74                 total_rss << (PAGE_SHIFT-10),
  75                 anon << (PAGE_SHIFT-10),
  76                 file << (PAGE_SHIFT-10),
  77                 shmem << (PAGE_SHIFT-10),
  78                 mm->data_vm << (PAGE_SHIFT-10),
  79                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  80                 ptes >> 10,
  81                 pmds >> 10,
  82                 swap << (PAGE_SHIFT-10));
  83         hugetlb_report_usage(m, mm);
  84 }
  85
  86 unsigned long task_vsize(struct mm_struct *mm)
  87 {
  88         return PAGE_SIZE * mm->total_vm;
  89 }
  90
  91 unsigned long task_statm(struct mm_struct *mm,
  92                          unsigned long *shared, unsigned long *text,
  93                          unsigned long *data, unsigned long *resident)
  94 {
  95         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  96                         get_mm_counter(mm, MM_SHMEMPAGES);
  97         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  98                                                                 >> PAGE_SHIFT;
  99         *data = mm->data_vm + mm->stack_vm;
 100         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
 101         return mm->total_vm;
 102 }
 103
 104 #ifdef CONFIG_NUMA
 105 /*
 106  * Save get_task_policy() for show_numa_map().
 107  */
 108 static void hold_task_mempolicy(struct proc_maps_private *priv)
 109 {
 110         struct task_struct *task = priv->task;
 111
 112         task_lock(task);
 113         priv->task_mempolicy = get_task_policy(task);
 114         mpol_get(priv->task_mempolicy);
 115         task_unlock(task);
 116 }
 117 static void release_task_mempolicy(struct proc_maps_private *priv)
 118 {
 119         mpol_put(priv->task_mempolicy);
 120 }
 121 #else
 122 static void hold_task_mempolicy(struct proc_maps_private *priv)
 123 {
 124 }
 125 static void release_task_mempolicy(struct proc_maps_private *priv)
 126 {
 127 }
 128 #endif
 129
 130 static void vma_stop(struct proc_maps_private *priv)
 131 {
 132         struct mm_struct *mm = priv->mm;
 133
 134         release_task_mempolicy(priv);
 135         up_read(&mm->mmap_sem);
 136         mmput(mm);
 137 }
 138
 139 static struct vm_area_struct *
 140 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
 141 {
 142         if (vma == priv->tail_vma)
 143                 return NULL;
 144         return vma->vm_next ?: priv->tail_vma;
 145 }
 146
 147 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
 148 {
 149         if (m->count < m->size) /* vma is copied successfully */
 150                 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
 151 }
 152
 153 static void *m_start(struct seq_file *m, loff_t *ppos)
 154 {
 155         struct proc_maps_private *priv = m->private;
 156         unsigned long last_addr = m->version;
 157         struct mm_struct *mm;
 158         struct vm_area_struct *vma;
 159         unsigned int pos = *ppos;
 160
 161         /* See m_cache_vma(). Zero at the start or after lseek. */
 162         if (last_addr == -1UL)
 163                 return NULL;
 164
 165         priv->task = get_proc_task(priv->inode);
 166         if (!priv->task)
 167                 return ERR_PTR(-ESRCH);
 168
 169         mm = priv->mm;
 170         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
 171                 return NULL;
 172
 173         down_read(&mm->mmap_sem);
 174         hold_task_mempolicy(priv);
 175         priv->tail_vma = get_gate_vma(mm);
 176
 177         if (last_addr) {
 178                 vma = find_vma(mm, last_addr - 1);
 179                 if (vma && vma->vm_start <= last_addr)
 180                         vma = m_next_vma(priv, vma);
 181                 if (vma)
 182                         return vma;
 183         }
 184
 185         m->version = 0;
 186         if (pos < mm->map_count) {
 187                 for (vma = mm->mmap; pos; pos--) {
 188                         m->version = vma->vm_start;
 189                         vma = vma->vm_next;
 190                 }
 191                 return vma;
 192         }
 193
 194         /* we do not bother to update m->version in this case */
 195         if (pos == mm->map_count && priv->tail_vma)
 196                 return priv->tail_vma;
 197
 198         vma_stop(priv);
 199         return NULL;
 200 }
 201
 202 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 203 {
 204         struct proc_maps_private *priv = m->private;
 205         struct vm_area_struct *next;
 206
 207         (*pos)++;
 208         next = m_next_vma(priv, v);
 209         if (!next)
 210                 vma_stop(priv);
 211         return next;
 212 }
 213
 214 static void m_stop(struct seq_file *m, void *v)
 215 {
 216         struct proc_maps_private *priv = m->private;
 217
 218         if (!IS_ERR_OR_NULL(v))
 219                 vma_stop(priv);
 220         if (priv->task) {
 221                 put_task_struct(priv->task);
 222                 priv->task = NULL;
 223         }
 224 }
 225
 226 static int proc_maps_open(struct inode *inode, struct file *file,
 227                         const struct seq_operations *ops, int psize)
 228 {
 229         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 230
 231         if (!priv)
 232                 return -ENOMEM;
 233
 234         priv->inode = inode;
 235         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 236         if (IS_ERR(priv->mm)) {
 237                 int err = PTR_ERR(priv->mm);
 238
 239                 seq_release_private(inode, file);
 240                 return err;
 241         }
 242
 243         return 0;
 244 }
 245
 246 static int proc_map_release(struct inode *inode, struct file *file)
 247 {
 248         struct seq_file *seq = file->private_data;
 249         struct proc_maps_private *priv = seq->private;
 250
 251         if (priv->mm)
 252                 mmdrop(priv->mm);
 253
 254         return seq_release_private(inode, file);
 255 }
 256
 257 static int do_maps_open(struct inode *inode, struct file *file,
 258                         const struct seq_operations *ops)
 259 {
 260         return proc_maps_open(inode, file, ops,
 261                                 sizeof(struct proc_maps_private));
 262 }
 263
 264 /*
 265  * Indicate if the VMA is a stack for the given task; for
 266  * /proc/PID/maps that is the stack of the main task.
 267  */
 268 static int is_stack(struct proc_maps_private *priv,
 269                     struct vm_area_struct *vma)
 270 {
 271         /*
 272          * We make no effort to guess what a given thread considers to be
 273          * its "stack".  It's not even well-defined for programs written
 274          * languages like Go.
 275          */
 276         return vma->vm_start <= vma->vm_mm->start_stack &&
 277                 vma->vm_end >= vma->vm_mm->start_stack;
 278 }
 279
 280 static void
 281 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 282 {
 283         struct mm_struct *mm = vma->vm_mm;
 284         struct file *file = vma->vm_file;
 285         struct proc_maps_private *priv = m->private;
 286         vm_flags_t flags = vma->vm_flags;
 287         unsigned long ino = 0;
 288         unsigned long long pgoff = 0;
 289         unsigned long start, end;
 290         dev_t dev = 0;
 291         const char *name = NULL;
 292
 293         if (file) {
 294                 struct inode *inode = file_inode(vma->vm_file);
 295                 dev = inode->i_sb->s_dev;
 296                 ino = inode->i_ino;
 297                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 298         }
 299
 300         /* We don't show the stack guard page in /proc/maps */
 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 anonymous_thp;
 440         unsigned long shmem_thp;
 441         unsigned long swap;
 442         unsigned long shared_hugetlb;
 443         unsigned long private_hugetlb;
 444         u64 pss;
 445         u64 swap_pss;
 446         bool check_shmem_swap;
 447 };
 448
 449 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 450                 bool compound, bool young, bool dirty)
 451 {
 452         int i, nr = compound ? 1 << compound_order(page) : 1;
 453         unsigned long size = nr * PAGE_SIZE;
 454
 455         if (PageAnon(page))
 456                 mss->anonymous += size;
 457
 458         mss->resident += size;
 459         /* Accumulate the size in pages that have been accessed. */
 460         if (young || page_is_young(page) || PageReferenced(page))
 461                 mss->referenced += size;
 462
 463         /*
 464          * page_count(page) == 1 guarantees the page is mapped exactly once.
 465          * If any subpage of the compound page mapped with PTE it would elevate
 466          * page_count().
 467          */
 468         if (page_count(page) == 1) {
 469                 if (dirty || PageDirty(page))
 470                         mss->private_dirty += size;
 471                 else
 472                         mss->private_clean += size;
 473                 mss->pss += (u64)size << PSS_SHIFT;
 474                 return;
 475         }
 476
 477         for (i = 0; i < nr; i++, page++) {
 478                 int mapcount = page_mapcount(page);
 479
 480                 if (mapcount >= 2) {
 481                         if (dirty || PageDirty(page))
 482                                 mss->shared_dirty += PAGE_SIZE;
 483                         else
 484                                 mss->shared_clean += PAGE_SIZE;
 485                         mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
 486                 } else {
 487                         if (dirty || PageDirty(page))
 488                                 mss->private_dirty += PAGE_SIZE;
 489                         else
 490                                 mss->private_clean += PAGE_SIZE;
 491                         mss->pss += PAGE_SIZE << PSS_SHIFT;
 492                 }
 493         }
 494 }
 495
 496 #ifdef CONFIG_SHMEM
 497 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 498                 struct mm_walk *walk)
 499 {
 500         struct mem_size_stats *mss = walk->private;
 501
 502         mss->swap += shmem_partial_swap_usage(
 503                         walk->vma->vm_file->f_mapping, addr, end);
 504
 505         return 0;
 506 }
 507 #endif
 508
 509 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 510                 struct mm_walk *walk)
 511 {
 512         struct mem_size_stats *mss = walk->private;
 513         struct vm_area_struct *vma = walk->vma;
 514         struct page *page = NULL;
 515
 516         if (pte_present(*pte)) {
 517                 page = vm_normal_page(vma, addr, *pte);
 518         } else if (is_swap_pte(*pte)) {
 519                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 520
 521                 if (!non_swap_entry(swpent)) {
 522                         int mapcount;
 523
 524                         mss->swap += PAGE_SIZE;
 525                         mapcount = swp_swapcount(swpent);
 526                         if (mapcount >= 2) {
 527                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 528
 529                                 do_div(pss_delta, mapcount);
 530                                 mss->swap_pss += pss_delta;
 531                         } else {
 532                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 533                         }
 534                 } else if (is_migration_entry(swpent))
 535                         page = migration_entry_to_page(swpent);
 536         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 537                                                         && pte_none(*pte))) {
 538                 page = find_get_entry(vma->vm_file->f_mapping,
 539                                                 linear_page_index(vma, addr));
 540                 if (!page)
 541                         return;
 542
 543                 if (radix_tree_exceptional_entry(page))
 544                         mss->swap += PAGE_SIZE;
 545                 else
 546                         put_page(page);
 547
 548                 return;
 549         }
 550
 551         if (!page)
 552                 return;
 553
 554         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
 555 }
 556
 557 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 558 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 559                 struct mm_walk *walk)
 560 {
 561         struct mem_size_stats *mss = walk->private;
 562         struct vm_area_struct *vma = walk->vma;
 563         struct page *page;
 564
 565         /* FOLL_DUMP will return -EFAULT on huge zero page */
 566         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 567         if (IS_ERR_OR_NULL(page))
 568                 return;
 569         if (PageAnon(page))
 570                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 571         else if (PageSwapBacked(page))
 572                 mss->shmem_thp += HPAGE_PMD_SIZE;
 573         else if (is_zone_device_page(page))
 574                 /* pass */;
 575         else
 576                 VM_BUG_ON_PAGE(1, page);
 577         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
 578 }
 579 #else
 580 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 581                 struct mm_walk *walk)
 582 {
 583 }
 584 #endif
 585
 586 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 587                            struct mm_walk *walk)
 588 {
 589         struct vm_area_struct *vma = walk->vma;
 590         pte_t *pte;
 591         spinlock_t *ptl;
 592
 593         ptl = pmd_trans_huge_lock(pmd, vma);
 594         if (ptl) {
 595                 smaps_pmd_entry(pmd, addr, walk);
 596                 spin_unlock(ptl);
 597                 return 0;
 598         }
 599
 600         if (pmd_trans_unstable(pmd))
 601                 return 0;
 602         /*
 603          * The mmap_sem held all the way back in m_start() is what
 604          * keeps khugepaged out of here and from collapsing things
 605          * in here.
 606          */
 607         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 608         for (; addr != end; pte++, addr += PAGE_SIZE)
 609                 smaps_pte_entry(pte, addr, walk);
 610         pte_unmap_unlock(pte - 1, ptl);
 611         cond_resched();
 612         return 0;
 613 }
 614
 615 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 616 {
 617         /*
 618          * Don't forget to update Documentation/ on changes.
 619          */
 620         static const char mnemonics[BITS_PER_LONG][2] = {
 621                 /*
 622                  * In case if we meet a flag we don't know about.
 623                  */
 624                 [0 ... (BITS_PER_LONG-1)] = "??",
 625
 626                 [ilog2(VM_READ)]        = "rd",
 627                 [ilog2(VM_WRITE)]       = "wr",
 628                 [ilog2(VM_EXEC)]        = "ex",
 629                 [ilog2(VM_SHARED)]      = "sh",
 630                 [ilog2(VM_MAYREAD)]     = "mr",
 631                 [ilog2(VM_MAYWRITE)]    = "mw",
 632                 [ilog2(VM_MAYEXEC)]     = "me",
 633                 [ilog2(VM_MAYSHARE)]    = "ms",
 634                 [ilog2(VM_GROWSDOWN)]   = "gd",
 635                 [ilog2(VM_PFNMAP)]      = "pf",
 636                 [ilog2(VM_DENYWRITE)]   = "dw",
 637 #ifdef CONFIG_X86_INTEL_MPX
 638                 [ilog2(VM_MPX)]         = "mp",
 639 #endif
 640                 [ilog2(VM_LOCKED)]      = "lo",
 641                 [ilog2(VM_IO)]          = "io",
 642                 [ilog2(VM_SEQ_READ)]    = "sr",
 643                 [ilog2(VM_RAND_READ)]   = "rr",
 644                 [ilog2(VM_DONTCOPY)]    = "dc",
 645                 [ilog2(VM_DONTEXPAND)]  = "de",
 646                 [ilog2(VM_ACCOUNT)]     = "ac",
 647                 [ilog2(VM_NORESERVE)]   = "nr",
 648                 [ilog2(VM_HUGETLB)]     = "ht",
 649                 [ilog2(VM_ARCH_1)]      = "ar",
 650                 [ilog2(VM_DONTDUMP)]    = "dd",
 651 #ifdef CONFIG_MEM_SOFT_DIRTY
 652                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 653 #endif
 654                 [ilog2(VM_MIXEDMAP)]    = "mm",
 655                 [ilog2(VM_HUGEPAGE)]    = "hg",
 656                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 657                 [ilog2(VM_MERGEABLE)]   = "mg",
 658                 [ilog2(VM_UFFD_MISSING)]= "um",
 659                 [ilog2(VM_UFFD_WP)]     = "uw",
 660 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 661                 /* These come out via ProtectionKey: */
 662                 [ilog2(VM_PKEY_BIT0)]   = "",
 663                 [ilog2(VM_PKEY_BIT1)]   = "",
 664                 [ilog2(VM_PKEY_BIT2)]   = "",
 665                 [ilog2(VM_PKEY_BIT3)]   = "",
 666 #endif
 667         };
 668         size_t i;
 669
 670         seq_puts(m, "VmFlags: ");
 671         for (i = 0; i < BITS_PER_LONG; i++) {
 672                 if (!mnemonics[i][0])
 673                         continue;
 674                 if (vma->vm_flags & (1UL << i)) {
 675                         seq_printf(m, "%c%c ",
 676                                    mnemonics[i][0], mnemonics[i][1]);
 677                 }
 678         }
 679         seq_putc(m, '\n');
 680 }
 681
 682 #ifdef CONFIG_HUGETLB_PAGE
 683 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 684                                  unsigned long addr, unsigned long end,
 685                                  struct mm_walk *walk)
 686 {
 687         struct mem_size_stats *mss = walk->private;
 688         struct vm_area_struct *vma = walk->vma;
 689         struct page *page = NULL;
 690
 691         if (pte_present(*pte)) {
 692                 page = vm_normal_page(vma, addr, *pte);
 693         } else if (is_swap_pte(*pte)) {
 694                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 695
 696                 if (is_migration_entry(swpent))
 697                         page = migration_entry_to_page(swpent);
 698         }
 699         if (page) {
 700                 int mapcount = page_mapcount(page);
 701
 702                 if (mapcount >= 2)
 703                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 704                 else
 705                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 706         }
 707         return 0;
 708 }
 709 #endif /* HUGETLB_PAGE */
 710
 711 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
 712 {
 713 }
 714
 715 static int show_smap(struct seq_file *m, void *v, int is_pid)
 716 {
 717         struct vm_area_struct *vma = v;
 718         struct mem_size_stats mss;
 719         struct mm_walk smaps_walk = {
 720                 .pmd_entry = smaps_pte_range,
 721 #ifdef CONFIG_HUGETLB_PAGE
 722                 .hugetlb_entry = smaps_hugetlb_range,
 723 #endif
 724                 .mm = vma->vm_mm,
 725                 .private = &mss,
 726         };
 727
 728         memset(&mss, 0, sizeof mss);
 729
 730 #ifdef CONFIG_SHMEM
 731         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 732                 /*
 733                  * For shared or readonly shmem mappings we know that all
 734                  * swapped out pages belong to the shmem object, and we can
 735                  * obtain the swap value much more efficiently. For private
 736                  * writable mappings, we might have COW pages that are
 737                  * not affected by the parent swapped out pages of the shmem
 738                  * object, so we have to distinguish them during the page walk.
 739                  * Unless we know that the shmem object (or the part mapped by
 740                  * our VMA) has no swapped out pages at all.
 741                  */
 742                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 743
 744                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 745                                         !(vma->vm_flags & VM_WRITE)) {
 746                         mss.swap = shmem_swapped;
 747                 } else {
 748                         mss.check_shmem_swap = true;
 749                         smaps_walk.pte_hole = smaps_pte_hole;
 750                 }
 751         }
 752 #endif
 753
 754         /* mmap_sem is held in m_start */
 755         walk_page_vma(vma, &smaps_walk);
 756
 757         show_map_vma(m, vma, is_pid);
 758
 759         seq_printf(m,
 760                    "Size:           %8lu kB\n"
 761                    "Rss:            %8lu kB\n"
 762                    "Pss:            %8lu kB\n"
 763                    "Shared_Clean:   %8lu kB\n"
 764                    "Shared_Dirty:   %8lu kB\n"
 765                    "Private_Clean:  %8lu kB\n"
 766                    "Private_Dirty:  %8lu kB\n"
 767                    "Referenced:     %8lu kB\n"
 768                    "Anonymous:      %8lu kB\n"
 769                    "AnonHugePages:  %8lu kB\n"
 770                    "ShmemPmdMapped: %8lu kB\n"
 771                    "Shared_Hugetlb: %8lu kB\n"
 772                    "Private_Hugetlb: %7lu kB\n"
 773                    "Swap:           %8lu kB\n"
 774                    "SwapPss:        %8lu kB\n"
 775                    "KernelPageSize: %8lu kB\n"
 776                    "MMUPageSize:    %8lu kB\n"
 777                    "Locked:         %8lu kB\n",
 778                    (vma->vm_end - vma->vm_start) >> 10,
 779                    mss.resident >> 10,
 780                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 781                    mss.shared_clean  >> 10,
 782                    mss.shared_dirty  >> 10,
 783                    mss.private_clean >> 10,
 784                    mss.private_dirty >> 10,
 785                    mss.referenced >> 10,
 786                    mss.anonymous >> 10,
 787                    mss.anonymous_thp >> 10,
 788                    mss.shmem_thp >> 10,
 789                    mss.shared_hugetlb >> 10,
 790                    mss.private_hugetlb >> 10,
 791                    mss.swap >> 10,
 792                    (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
 793                    vma_kernel_pagesize(vma) >> 10,
 794                    vma_mmu_pagesize(vma) >> 10,
 795                    (vma->vm_flags & VM_LOCKED) ?
 796                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 797
 798         arch_show_smap(m, vma);
 799         show_smap_vma_flags(m, vma);
 800         m_cache_vma(m, vma);
 801         return 0;
 802 }
 803
 804 static int show_pid_smap(struct seq_file *m, void *v)
 805 {
 806         return show_smap(m, v, 1);
 807 }
 808
 809 static int show_tid_smap(struct seq_file *m, void *v)
 810 {
 811         return show_smap(m, v, 0);
 812 }
 813
 814 static const struct seq_operations proc_pid_smaps_op = {
 815         .start  = m_start,
 816         .next   = m_next,
 817         .stop   = m_stop,
 818         .show   = show_pid_smap
 819 };
 820
 821 static const struct seq_operations proc_tid_smaps_op = {
 822         .start  = m_start,
 823         .next   = m_next,
 824         .stop   = m_stop,
 825         .show   = show_tid_smap
 826 };
 827
 828 static int pid_smaps_open(struct inode *inode, struct file *file)
 829 {
 830         return do_maps_open(inode, file, &proc_pid_smaps_op);
 831 }
 832
 833 static int tid_smaps_open(struct inode *inode, struct file *file)
 834 {
 835         return do_maps_open(inode, file, &proc_tid_smaps_op);
 836 }
 837
 838 const struct file_operations proc_pid_smaps_operations = {
 839         .open           = pid_smaps_open,
 840         .read           = seq_read,
 841         .llseek         = seq_lseek,
 842         .release        = proc_map_release,
 843 };
 844
 845 const struct file_operations proc_tid_smaps_operations = {
 846         .open           = tid_smaps_open,
 847         .read           = seq_read,
 848         .llseek         = seq_lseek,
 849         .release        = proc_map_release,
 850 };
 851
 852 enum clear_refs_types {
 853         CLEAR_REFS_ALL = 1,
 854         CLEAR_REFS_ANON,
 855         CLEAR_REFS_MAPPED,
 856         CLEAR_REFS_SOFT_DIRTY,
 857         CLEAR_REFS_MM_HIWATER_RSS,
 858         CLEAR_REFS_LAST,
 859 };
 860
 861 struct clear_refs_private {
 862         enum clear_refs_types type;
 863 };
 864
 865 #ifdef CONFIG_MEM_SOFT_DIRTY
 866 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 867                 unsigned long addr, pte_t *pte)
 868 {
 869         /*
 870          * The soft-dirty tracker uses #PF-s to catch writes
 871          * to pages, so write-protect the pte as well. See the
 872          * Documentation/vm/soft-dirty.txt for full description
 873          * of how soft-dirty works.
 874          */
 875         pte_t ptent = *pte;
 876
 877         if (pte_present(ptent)) {
 878                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 879                 ptent = pte_wrprotect(ptent);
 880                 ptent = pte_clear_soft_dirty(ptent);
 881                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 882         } else if (is_swap_pte(ptent)) {
 883                 ptent = pte_swp_clear_soft_dirty(ptent);
 884                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 885         }
 886 }
 887 #else
 888 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 889                 unsigned long addr, pte_t *pte)
 890 {
 891 }
 892 #endif
 893
 894 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 895 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 896                 unsigned long addr, pmd_t *pmdp)
 897 {
 898         pmd_t pmd = *pmdp;
 899
 900         /* See comment in change_huge_pmd() */
 901         pmdp_invalidate(vma, addr, pmdp);
 902         if (pmd_dirty(*pmdp))
 903                 pmd = pmd_mkdirty(pmd);
 904         if (pmd_young(*pmdp))
 905                 pmd = pmd_mkyoung(pmd);
 906
 907         pmd = pmd_wrprotect(pmd);
 908         pmd = pmd_clear_soft_dirty(pmd);
 909
 910         set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 911 }
 912 #else
 913 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 914                 unsigned long addr, pmd_t *pmdp)
 915 {
 916 }
 917 #endif
 918
 919 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 920                                 unsigned long end, struct mm_walk *walk)
 921 {
 922         struct clear_refs_private *cp = walk->private;
 923         struct vm_area_struct *vma = walk->vma;
 924         pte_t *pte, ptent;
 925         spinlock_t *ptl;
 926         struct page *page;
 927
 928         ptl = pmd_trans_huge_lock(pmd, vma);
 929         if (ptl) {
 930                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 931                         clear_soft_dirty_pmd(vma, addr, pmd);
 932                         goto out;
 933                 }
 934
 935                 page = pmd_page(*pmd);
 936
 937                 /* Clear accessed and referenced bits. */
 938                 pmdp_test_and_clear_young(vma, addr, pmd);
 939                 test_and_clear_page_young(page);
 940                 ClearPageReferenced(page);
 941 out:
 942                 spin_unlock(ptl);
 943                 return 0;
 944         }
 945
 946         if (pmd_trans_unstable(pmd))
 947                 return 0;
 948
 949         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 950         for (; addr != end; pte++, addr += PAGE_SIZE) {
 951                 ptent = *pte;
 952
 953                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 954                         clear_soft_dirty(vma, addr, pte);
 955                         continue;
 956                 }
 957
 958                 if (!pte_present(ptent))
 959                         continue;
 960
 961                 page = vm_normal_page(vma, addr, ptent);
 962                 if (!page)
 963                         continue;
 964
 965                 /* Clear accessed and referenced bits. */
 966                 ptep_test_and_clear_young(vma, addr, pte);
 967                 test_and_clear_page_young(page);
 968                 ClearPageReferenced(page);
 969         }
 970         pte_unmap_unlock(pte - 1, ptl);
 971         cond_resched();
 972         return 0;
 973 }
 974
 975 static int clear_refs_test_walk(unsigned long start, unsigned long end,
 976                                 struct mm_walk *walk)
 977 {
 978         struct clear_refs_private *cp = walk->private;
 979         struct vm_area_struct *vma = walk->vma;
 980
 981         if (vma->vm_flags & VM_PFNMAP)
 982                 return 1;
 983
 984         /*
 985          * Writing 1 to /proc/pid/clear_refs affects all pages.
 986          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
 987          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
 988          * Writing 4 to /proc/pid/clear_refs affects all pages.
 989          */
 990         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
 991                 return 1;
 992         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
 993                 return 1;
 994         return 0;
 995 }
 996
 997 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 998                                 size_t count, loff_t *ppos)
 999 {
1000         struct task_struct *task;
1001         char buffer[PROC_NUMBUF];
1002         struct mm_struct *mm;
1003         struct vm_area_struct *vma;
1004         enum clear_refs_types type;
1005         int itype;
1006         int rv;
1007
1008         memset(buffer, 0, sizeof(buffer));
1009         if (count > sizeof(buffer) - 1)
1010                 count = sizeof(buffer) - 1;
1011         if (copy_from_user(buffer, buf, count))
1012                 return -EFAULT;
1013         rv = kstrtoint(strstrip(buffer), 10, &itype);
1014         if (rv < 0)
1015                 return rv;
1016         type = (enum clear_refs_types)itype;
1017         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1018                 return -EINVAL;
1019
1020         task = get_proc_task(file_inode(file));
1021         if (!task)
1022                 return -ESRCH;
1023         mm = get_task_mm(task);
1024         if (mm) {
1025                 struct clear_refs_private cp = {
1026                         .type = type,
1027                 };
1028                 struct mm_walk clear_refs_walk = {
1029                         .pmd_entry = clear_refs_pte_range,
1030                         .test_walk = clear_refs_test_walk,
1031                         .mm = mm,
1032                         .private = &cp,
1033                 };
1034
1035                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1036                         if (down_write_killable(&mm->mmap_sem)) {
1037                                 count = -EINTR;
1038                                 goto out_mm;
1039                         }
1040
1041                         /*
1042                          * Writing 5 to /proc/pid/clear_refs resets the peak
1043                          * resident set size to this mm's current rss value.
1044                          */
1045                         reset_mm_hiwater_rss(mm);
1046                         up_write(&mm->mmap_sem);
1047                         goto out_mm;
1048                 }
1049
1050                 down_read(&mm->mmap_sem);
1051                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1052                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1053                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1054                                         continue;
1055                                 up_read(&mm->mmap_sem);
1056                                 if (down_write_killable(&mm->mmap_sem)) {
1057                                         count = -EINTR;
1058                                         goto out_mm;
1059                                 }
1060                                 /*
1061                                  * Avoid to modify vma->vm_flags
1062                                  * without locked ops while the
1063                                  * coredump reads the vm_flags.
1064                                  */
1065                                 if (!mmget_still_valid(mm)) {
1066                                         /*
1067                                          * Silently return "count"
1068                                          * like if get_task_mm()
1069                                          * failed. FIXME: should this
1070                                          * function have returned
1071                                          * -ESRCH if get_task_mm()
1072                                          * failed like if
1073                                          * get_proc_task() fails?
1074                                          */
1075                                         up_write(&mm->mmap_sem);
1076                                         goto out_mm;
1077                                 }
1078                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1079                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1080                                         vma_set_page_prot(vma);
1081                                 }
1082                                 downgrade_write(&mm->mmap_sem);
1083                                 break;
1084                         }
1085                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1086                 }
1087                 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1088                 if (type == CLEAR_REFS_SOFT_DIRTY)
1089                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1090                 flush_tlb_mm(mm);
1091                 up_read(&mm->mmap_sem);
1092 out_mm:
1093                 mmput(mm);
1094         }
1095         put_task_struct(task);
1096
1097         return count;
1098 }
1099
1100 const struct file_operations proc_clear_refs_operations = {
1101         .write          = clear_refs_write,
1102         .llseek         = noop_llseek,
1103 };
1104
1105 typedef struct {
1106         u64 pme;
1107 } pagemap_entry_t;
1108
1109 struct pagemapread {
1110         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1111         pagemap_entry_t *buffer;
1112         bool show_pfn;
1113 };
1114
1115 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1116 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1117
1118 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1119 #define PM_PFRAME_BITS          55
1120 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1121 #define PM_SOFT_DIRTY           BIT_ULL(55)
1122 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1123 #define PM_FILE                 BIT_ULL(61)
1124 #define PM_SWAP                 BIT_ULL(62)
1125 #define PM_PRESENT              BIT_ULL(63)
1126
1127 #define PM_END_OF_BUFFER    1
1128
1129 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1130 {
1131         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1132 }
1133
1134 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1135                           struct pagemapread *pm)
1136 {
1137         pm->buffer[pm->pos++] = *pme;
1138         if (pm->pos >= pm->len)
1139                 return PM_END_OF_BUFFER;
1140         return 0;
1141 }
1142
1143 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1144                                 struct mm_walk *walk)
1145 {
1146         struct pagemapread *pm = walk->private;
1147         unsigned long addr = start;
1148         int err = 0;
1149
1150         while (addr < end) {
1151                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1152                 pagemap_entry_t pme = make_pme(0, 0);
1153                 /* End of address space hole, which we mark as non-present. */
1154                 unsigned long hole_end;
1155
1156                 if (vma)
1157                         hole_end = min(end, vma->vm_start);
1158                 else
1159                         hole_end = end;
1160
1161                 for (; addr < hole_end; addr += PAGE_SIZE) {
1162                         err = add_to_pagemap(addr, &pme, pm);
1163                         if (err)
1164                                 goto out;
1165                 }
1166
1167                 if (!vma)
1168                         break;
1169
1170                 /* Addresses in the VMA. */
1171                 if (vma->vm_flags & VM_SOFTDIRTY)
1172                         pme = make_pme(0, PM_SOFT_DIRTY);
1173                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1174                         err = add_to_pagemap(addr, &pme, pm);
1175                         if (err)
1176                                 goto out;
1177                 }
1178         }
1179 out:
1180         return err;
1181 }
1182
1183 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1184                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1185 {
1186         u64 frame = 0, flags = 0;
1187         struct page *page = NULL;
1188
1189         if (pte_present(pte)) {
1190                 if (pm->show_pfn)
1191                         frame = pte_pfn(pte);
1192                 flags |= PM_PRESENT;
1193                 page = vm_normal_page(vma, addr, pte);
1194                 if (pte_soft_dirty(pte))
1195                         flags |= PM_SOFT_DIRTY;
1196         } else if (is_swap_pte(pte)) {
1197                 swp_entry_t entry;
1198                 if (pte_swp_soft_dirty(pte))
1199                         flags |= PM_SOFT_DIRTY;
1200                 entry = pte_to_swp_entry(pte);
1201                 frame = swp_type(entry) |
1202                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1203                 flags |= PM_SWAP;
1204                 if (is_migration_entry(entry))
1205                         page = migration_entry_to_page(entry);
1206         }
1207
1208         if (page && !PageAnon(page))
1209                 flags |= PM_FILE;
1210         if (page && page_mapcount(page) == 1)
1211                 flags |= PM_MMAP_EXCLUSIVE;
1212         if (vma->vm_flags & VM_SOFTDIRTY)
1213                 flags |= PM_SOFT_DIRTY;
1214
1215         return make_pme(frame, flags);
1216 }
1217
1218 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1219                              struct mm_walk *walk)
1220 {
1221         struct vm_area_struct *vma = walk->vma;
1222         struct pagemapread *pm = walk->private;
1223         spinlock_t *ptl;
1224         pte_t *pte, *orig_pte;
1225         int err = 0;
1226
1227 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1228         ptl = pmd_trans_huge_lock(pmdp, vma);
1229         if (ptl) {
1230                 u64 flags = 0, frame = 0;
1231                 pmd_t pmd = *pmdp;
1232
1233                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1234                         flags |= PM_SOFT_DIRTY;
1235
1236                 /*
1237                  * Currently pmd for thp is always present because thp
1238                  * can not be swapped-out, migrated, or HWPOISONed
1239                  * (split in such cases instead.)
1240                  * This if-check is just to prepare for future implementation.
1241                  */
1242                 if (pmd_present(pmd)) {
1243                         struct page *page = pmd_page(pmd);
1244
1245                         if (page_mapcount(page) == 1)
1246                                 flags |= PM_MMAP_EXCLUSIVE;
1247
1248                         flags |= PM_PRESENT;
1249                         if (pm->show_pfn)
1250                                 frame = pmd_pfn(pmd) +
1251                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1252                 }
1253
1254                 for (; addr != end; addr += PAGE_SIZE) {
1255                         pagemap_entry_t pme = make_pme(frame, flags);
1256
1257                         err = add_to_pagemap(addr, &pme, pm);
1258                         if (err)
1259                                 break;
1260                         if (pm->show_pfn && (flags & PM_PRESENT))
1261                                 frame++;
1262                 }
1263                 spin_unlock(ptl);
1264                 return err;
1265         }
1266
1267         if (pmd_trans_unstable(pmdp))
1268                 return 0;
1269 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1270
1271         /*
1272          * We can assume that @vma always points to a valid one and @end never
1273          * goes beyond vma->vm_end.
1274          */
1275         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1276         for (; addr < end; pte++, addr += PAGE_SIZE) {
1277                 pagemap_entry_t pme;
1278
1279                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1280                 err = add_to_pagemap(addr, &pme, pm);
1281                 if (err)
1282                         break;
1283         }
1284         pte_unmap_unlock(orig_pte, ptl);
1285
1286         cond_resched();
1287
1288         return err;
1289 }
1290
1291 #ifdef CONFIG_HUGETLB_PAGE
1292 /* This function walks within one hugetlb entry in the single call */
1293 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1294                                  unsigned long addr, unsigned long end,
1295                                  struct mm_walk *walk)
1296 {
1297         struct pagemapread *pm = walk->private;
1298         struct vm_area_struct *vma = walk->vma;
1299         u64 flags = 0, frame = 0;
1300         int err = 0;
1301         pte_t pte;
1302
1303         if (vma->vm_flags & VM_SOFTDIRTY)
1304                 flags |= PM_SOFT_DIRTY;
1305
1306         pte = huge_ptep_get(ptep);
1307         if (pte_present(pte)) {
1308                 struct page *page = pte_page(pte);
1309
1310                 if (!PageAnon(page))
1311                         flags |= PM_FILE;
1312
1313                 if (page_mapcount(page) == 1)
1314                         flags |= PM_MMAP_EXCLUSIVE;
1315
1316                 flags |= PM_PRESENT;
1317                 if (pm->show_pfn)
1318                         frame = pte_pfn(pte) +
1319                                 ((addr & ~hmask) >> PAGE_SHIFT);
1320         }
1321
1322         for (; addr != end; addr += PAGE_SIZE) {
1323                 pagemap_entry_t pme = make_pme(frame, flags);
1324
1325                 err = add_to_pagemap(addr, &pme, pm);
1326                 if (err)
1327                         return err;
1328                 if (pm->show_pfn && (flags & PM_PRESENT))
1329                         frame++;
1330         }
1331
1332         cond_resched();
1333
1334         return err;
1335 }
1336 #endif /* HUGETLB_PAGE */
1337
1338 /*
1339  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1340  *
1341  * For each page in the address space, this file contains one 64-bit entry
1342  * consisting of the following:
1343  *
1344  * Bits 0-54  page frame number (PFN) if present
1345  * Bits 0-4   swap type if swapped
1346  * Bits 5-54  swap offset if swapped
1347  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1348  * Bit  56    page exclusively mapped
1349  * Bits 57-60 zero
1350  * Bit  61    page is file-page or shared-anon
1351  * Bit  62    page swapped
1352  * Bit  63    page present
1353  *
1354  * If the page is not present but in swap, then the PFN contains an
1355  * encoding of the swap file number and the page's offset into the
1356  * swap. Unmapped pages return a null PFN. This allows determining
1357  * precisely which pages are mapped (or in swap) and comparing mapped
1358  * pages between processes.
1359  *
1360  * Efficient users of this interface will use /proc/pid/maps to
1361  * determine which areas of memory are actually mapped and llseek to
1362  * skip over unmapped regions.
1363  */
1364 static ssize_t pagemap_read(struct file *file, char __user *buf,
1365                             size_t count, loff_t *ppos)
1366 {
1367         struct mm_struct *mm = file->private_data;
1368         struct pagemapread pm;
1369         struct mm_walk pagemap_walk = {};
1370         unsigned long src;
1371         unsigned long svpfn;
1372         unsigned long start_vaddr;
1373         unsigned long end_vaddr;
1374         int ret = 0, copied = 0;
1375
1376         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1377                 goto out;
1378
1379         ret = -EINVAL;
1380         /* file position must be aligned */
1381         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1382                 goto out_mm;
1383
1384         ret = 0;
1385         if (!count)
1386                 goto out_mm;
1387
1388         /* do not disclose physical addresses: attack vector */
1389         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1390
1391         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1392         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1393         ret = -ENOMEM;
1394         if (!pm.buffer)
1395                 goto out_mm;
1396
1397         pagemap_walk.pmd_entry = pagemap_pmd_range;
1398         pagemap_walk.pte_hole = pagemap_pte_hole;
1399 #ifdef CONFIG_HUGETLB_PAGE
1400         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1401 #endif
1402         pagemap_walk.mm = mm;
1403         pagemap_walk.private = &pm;
1404
1405         src = *ppos;
1406         svpfn = src / PM_ENTRY_BYTES;
1407         start_vaddr = svpfn << PAGE_SHIFT;
1408         end_vaddr = mm->task_size;
1409
1410         /* watch out for wraparound */
1411         if (svpfn > mm->task_size >> PAGE_SHIFT)
1412                 start_vaddr = end_vaddr;
1413
1414         /*
1415          * The odds are that this will stop walking way
1416          * before end_vaddr, because the length of the
1417          * user buffer is tracked in "pm", and the walk
1418          * will stop when we hit the end of the buffer.
1419          */
1420         ret = 0;
1421         while (count && (start_vaddr < end_vaddr)) {
1422                 int len;
1423                 unsigned long end;
1424
1425                 pm.pos = 0;
1426                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1427                 /* overflow ? */
1428                 if (end < start_vaddr || end > end_vaddr)
1429                         end = end_vaddr;
1430                 down_read(&mm->mmap_sem);
1431                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1432                 up_read(&mm->mmap_sem);
1433                 start_vaddr = end;
1434
1435                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1436                 if (copy_to_user(buf, pm.buffer, len)) {
1437                         ret = -EFAULT;
1438                         goto out_free;
1439                 }
1440                 copied += len;
1441                 buf += len;
1442                 count -= len;
1443         }
1444         *ppos += copied;
1445         if (!ret || ret == PM_END_OF_BUFFER)
1446                 ret = copied;
1447
1448 out_free:
1449         kfree(pm.buffer);
1450 out_mm:
1451         mmput(mm);
1452 out:
1453         return ret;
1454 }
1455
1456 static int pagemap_open(struct inode *inode, struct file *file)
1457 {
1458         struct mm_struct *mm;
1459
1460         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1461         if (IS_ERR(mm))
1462                 return PTR_ERR(mm);
1463         file->private_data = mm;
1464         return 0;
1465 }
1466
1467 static int pagemap_release(struct inode *inode, struct file *file)
1468 {
1469         struct mm_struct *mm = file->private_data;
1470
1471         if (mm)
1472                 mmdrop(mm);
1473         return 0;
1474 }
1475
1476 const struct file_operations proc_pagemap_operations = {
1477         .llseek         = mem_lseek, /* borrow this */
1478         .read           = pagemap_read,
1479         .open           = pagemap_open,
1480         .release        = pagemap_release,
1481 };
1482 #endif /* CONFIG_PROC_PAGE_MONITOR */
1483
1484 #ifdef CONFIG_NUMA
1485
1486 struct numa_maps {
1487         unsigned long pages;
1488         unsigned long anon;
1489         unsigned long active;
1490         unsigned long writeback;
1491         unsigned long mapcount_max;
1492         unsigned long dirty;
1493         unsigned long swapcache;
1494         unsigned long node[MAX_NUMNODES];
1495 };
1496
1497 struct numa_maps_private {
1498         struct proc_maps_private proc_maps;
1499         struct numa_maps md;
1500 };
1501
1502 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1503                         unsigned long nr_pages)
1504 {
1505         int count = page_mapcount(page);
1506
1507         md->pages += nr_pages;
1508         if (pte_dirty || PageDirty(page))
1509                 md->dirty += nr_pages;
1510
1511         if (PageSwapCache(page))
1512                 md->swapcache += nr_pages;
1513
1514         if (PageActive(page) || PageUnevictable(page))
1515                 md->active += nr_pages;
1516
1517         if (PageWriteback(page))
1518                 md->writeback += nr_pages;
1519
1520         if (PageAnon(page))
1521                 md->anon += nr_pages;
1522
1523         if (count > md->mapcount_max)
1524                 md->mapcount_max = count;
1525
1526         md->node[page_to_nid(page)] += nr_pages;
1527 }
1528
1529 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1530                 unsigned long addr)
1531 {
1532         struct page *page;
1533         int nid;
1534
1535         if (!pte_present(pte))
1536                 return NULL;
1537
1538         page = vm_normal_page(vma, addr, pte);
1539         if (!page)
1540                 return NULL;
1541
1542         if (PageReserved(page))
1543                 return NULL;
1544
1545         nid = page_to_nid(page);
1546         if (!node_isset(nid, node_states[N_MEMORY]))
1547                 return NULL;
1548
1549         return page;
1550 }
1551
1552 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1553 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1554                                               struct vm_area_struct *vma,
1555                                               unsigned long addr)
1556 {
1557         struct page *page;
1558         int nid;
1559
1560         if (!pmd_present(pmd))
1561                 return NULL;
1562
1563         page = vm_normal_page_pmd(vma, addr, pmd);
1564         if (!page)
1565                 return NULL;
1566
1567         if (PageReserved(page))
1568                 return NULL;
1569
1570         nid = page_to_nid(page);
1571         if (!node_isset(nid, node_states[N_MEMORY]))
1572                 return NULL;
1573
1574         return page;
1575 }
1576 #endif
1577
1578 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1579                 unsigned long end, struct mm_walk *walk)
1580 {
1581         struct numa_maps *md = walk->private;
1582         struct vm_area_struct *vma = walk->vma;
1583         spinlock_t *ptl;
1584         pte_t *orig_pte;
1585         pte_t *pte;
1586
1587 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1588         ptl = pmd_trans_huge_lock(pmd, vma);
1589         if (ptl) {
1590                 struct page *page;
1591
1592                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1593                 if (page)
1594                         gather_stats(page, md, pmd_dirty(*pmd),
1595                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1596                 spin_unlock(ptl);
1597                 return 0;
1598         }
1599
1600         if (pmd_trans_unstable(pmd))
1601                 return 0;
1602 #endif
1603         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1604         do {
1605                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1606                 if (!page)
1607                         continue;
1608                 gather_stats(page, md, pte_dirty(*pte), 1);
1609
1610         } while (pte++, addr += PAGE_SIZE, addr != end);
1611         pte_unmap_unlock(orig_pte, ptl);
1612         cond_resched();
1613         return 0;
1614 }
1615 #ifdef CONFIG_HUGETLB_PAGE
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         pte_t huge_pte = huge_ptep_get(pte);
1620         struct numa_maps *md;
1621         struct page *page;
1622
1623         if (!pte_present(huge_pte))
1624                 return 0;
1625
1626         page = pte_page(huge_pte);
1627         if (!page)
1628                 return 0;
1629
1630         md = walk->private;
1631         gather_stats(page, md, pte_dirty(huge_pte), 1);
1632         return 0;
1633 }
1634
1635 #else
1636 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1637                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1638 {
1639         return 0;
1640 }
1641 #endif
1642
1643 /*
1644  * Display pages allocated per node and memory policy via /proc.
1645  */
1646 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1647 {
1648         struct numa_maps_private *numa_priv = m->private;
1649         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1650         struct vm_area_struct *vma = v;
1651         struct numa_maps *md = &numa_priv->md;
1652         struct file *file = vma->vm_file;
1653         struct mm_struct *mm = vma->vm_mm;
1654         struct mm_walk walk = {
1655                 .hugetlb_entry = gather_hugetlb_stats,
1656                 .pmd_entry = gather_pte_stats,
1657                 .private = md,
1658                 .mm = mm,
1659         };
1660         struct mempolicy *pol;
1661         char buffer[64];
1662         int nid;
1663
1664         if (!mm)
1665                 return 0;
1666
1667         /* Ensure we start with an empty set of numa_maps statistics. */
1668         memset(md, 0, sizeof(*md));
1669
1670         pol = __get_vma_policy(vma, vma->vm_start);
1671         if (pol) {
1672                 mpol_to_str(buffer, sizeof(buffer), pol);
1673                 mpol_cond_put(pol);
1674         } else {
1675                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1676         }
1677
1678         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1679
1680         if (file) {
1681                 seq_puts(m, " file=");
1682                 seq_file_path(m, file, "\n\t= ");
1683         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1684                 seq_puts(m, " heap");
1685         } else if (is_stack(proc_priv, vma)) {
1686                 seq_puts(m, " stack");
1687         }
1688
1689         if (is_vm_hugetlb_page(vma))
1690                 seq_puts(m, " huge");
1691
1692         /* mmap_sem is held by m_start */
1693         walk_page_vma(vma, &walk);
1694
1695         if (!md->pages)
1696                 goto out;
1697
1698         if (md->anon)
1699                 seq_printf(m, " anon=%lu", md->anon);
1700
1701         if (md->dirty)
1702                 seq_printf(m, " dirty=%lu", md->dirty);
1703
1704         if (md->pages != md->anon && md->pages != md->dirty)
1705                 seq_printf(m, " mapped=%lu", md->pages);
1706
1707         if (md->mapcount_max > 1)
1708                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1709
1710         if (md->swapcache)
1711                 seq_printf(m, " swapcache=%lu", md->swapcache);
1712
1713         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1714                 seq_printf(m, " active=%lu", md->active);
1715
1716         if (md->writeback)
1717                 seq_printf(m, " writeback=%lu", md->writeback);
1718
1719         for_each_node_state(nid, N_MEMORY)
1720                 if (md->node[nid])
1721                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1722
1723         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1724 out:
1725         seq_putc(m, '\n');
1726         m_cache_vma(m, vma);
1727         return 0;
1728 }
1729
1730 static int show_pid_numa_map(struct seq_file *m, void *v)
1731 {
1732         return show_numa_map(m, v, 1);
1733 }
1734
1735 static int show_tid_numa_map(struct seq_file *m, void *v)
1736 {
1737         return show_numa_map(m, v, 0);
1738 }
1739
1740 static const struct seq_operations proc_pid_numa_maps_op = {
1741         .start  = m_start,
1742         .next   = m_next,
1743         .stop   = m_stop,
1744         .show   = show_pid_numa_map,
1745 };
1746
1747 static const struct seq_operations proc_tid_numa_maps_op = {
1748         .start  = m_start,
1749         .next   = m_next,
1750         .stop   = m_stop,
1751         .show   = show_tid_numa_map,
1752 };
1753
1754 static int numa_maps_open(struct inode *inode, struct file *file,
1755                           const struct seq_operations *ops)
1756 {
1757         return proc_maps_open(inode, file, ops,
1758                                 sizeof(struct numa_maps_private));
1759 }
1760
1761 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1762 {
1763         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1764 }
1765
1766 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1767 {
1768         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1769 }
1770
1771 const struct file_operations proc_pid_numa_maps_operations = {
1772         .open           = pid_numa_maps_open,
1773         .read           = seq_read,
1774         .llseek         = seq_lseek,
1775         .release        = proc_map_release,
1776 };
1777
1778 const struct file_operations proc_tid_numa_maps_operations = {
1779         .open           = tid_numa_maps_open,
1780         .read           = seq_read,
1781         .llseek         = seq_lseek,
1782         .release        = proc_map_release,
1783 };
1784 #endif /* CONFIG_NUMA */