fs/proc/task_mmu.c

   1 #include <linux/mm.h>
   2 #include <linux/vmacache.h>
   3 #include <linux/hugetlb.h>
   4 #include <linux/huge_mm.h>
   5 #include <linux/mount.h>
   6 #include <linux/seq_file.h>
   7 #include <linux/highmem.h>
   8 #include <linux/ptrace.h>
   9 #include <linux/slab.h>
  10 #include <linux/pagemap.h>
  11 #include <linux/mempolicy.h>
  12 #include <linux/rmap.h>
  13 #include <linux/swap.h>
  14 #include <linux/swapops.h>
  15 #include <linux/mmu_notifier.h>
  16 #include <linux/page_idle.h>
  17
  18 #include <asm/elf.h>
  19 #include <asm/uaccess.h>
  20 #include <asm/tlbflush.h>
  21 #include "internal.h"
  22
  23 void task_mem(struct seq_file *m, struct mm_struct *mm)
  24 {
  25         unsigned long data, text, lib, swap, ptes, pmds;
  26         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  27
  28         /*
  29          * Note: to minimize their overhead, mm maintains hiwater_vm and
  30          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  31          * collector of these hiwater stats must therefore get total_vm
  32          * and rss too, which will usually be the higher.  Barriers? not
  33          * worth the effort, such snapshots can always be inconsistent.
  34          */
  35         hiwater_vm = total_vm = mm->total_vm;
  36         if (hiwater_vm < mm->hiwater_vm)
  37                 hiwater_vm = mm->hiwater_vm;
  38         hiwater_rss = total_rss = get_mm_rss(mm);
  39         if (hiwater_rss < mm->hiwater_rss)
  40                 hiwater_rss = mm->hiwater_rss;
  41
  42         data = mm->total_vm - mm->shared_vm - mm->stack_vm;
  43         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  44         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  45         swap = get_mm_counter(mm, MM_SWAPENTS);
  46         ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
  47         pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
  48         seq_printf(m,
  49                 "VmPeak:\t%8lu kB\n"
  50                 "VmSize:\t%8lu kB\n"
  51                 "VmLck:\t%8lu kB\n"
  52                 "VmPin:\t%8lu kB\n"
  53                 "VmHWM:\t%8lu kB\n"
  54                 "VmRSS:\t%8lu kB\n"
  55                 "VmData:\t%8lu kB\n"
  56                 "VmStk:\t%8lu kB\n"
  57                 "VmExe:\t%8lu kB\n"
  58                 "VmLib:\t%8lu kB\n"
  59                 "VmPTE:\t%8lu kB\n"
  60                 "VmPMD:\t%8lu kB\n"
  61                 "VmSwap:\t%8lu kB\n",
  62                 hiwater_vm << (PAGE_SHIFT-10),
  63                 total_vm << (PAGE_SHIFT-10),
  64                 mm->locked_vm << (PAGE_SHIFT-10),
  65                 mm->pinned_vm << (PAGE_SHIFT-10),
  66                 hiwater_rss << (PAGE_SHIFT-10),
  67                 total_rss << (PAGE_SHIFT-10),
  68                 data << (PAGE_SHIFT-10),
  69                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  70                 ptes >> 10,
  71                 pmds >> 10,
  72                 swap << (PAGE_SHIFT-10));
  73         hugetlb_report_usage(m, mm);
  74 }
  75
  76 unsigned long task_vsize(struct mm_struct *mm)
  77 {
  78         return PAGE_SIZE * mm->total_vm;
  79 }
  80
  81 unsigned long task_statm(struct mm_struct *mm,
  82                          unsigned long *shared, unsigned long *text,
  83                          unsigned long *data, unsigned long *resident)
  84 {
  85         *shared = get_mm_counter(mm, MM_FILEPAGES);
  86         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  87                                                                 >> PAGE_SHIFT;
  88         *data = mm->total_vm - mm->shared_vm;
  89         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
  90         return mm->total_vm;
  91 }
  92
  93 #ifdef CONFIG_NUMA
  94 /*
  95  * Save get_task_policy() for show_numa_map().
  96  */
  97 static void hold_task_mempolicy(struct proc_maps_private *priv)
  98 {
  99         struct task_struct *task = priv->task;
 100
 101         task_lock(task);
 102         priv->task_mempolicy = get_task_policy(task);
 103         mpol_get(priv->task_mempolicy);
 104         task_unlock(task);
 105 }
 106 static void release_task_mempolicy(struct proc_maps_private *priv)
 107 {
 108         mpol_put(priv->task_mempolicy);
 109 }
 110 #else
 111 static void hold_task_mempolicy(struct proc_maps_private *priv)
 112 {
 113 }
 114 static void release_task_mempolicy(struct proc_maps_private *priv)
 115 {
 116 }
 117 #endif
 118
 119 static void seq_print_vma_name(struct seq_file *m, struct vm_area_struct *vma)
 120 {
 121         const char __user *name = vma_get_anon_name(vma);
 122         struct mm_struct *mm = vma->vm_mm;
 123
 124         unsigned long page_start_vaddr;
 125         unsigned long page_offset;
 126         unsigned long num_pages;
 127         unsigned long max_len = NAME_MAX;
 128         int i;
 129
 130         page_start_vaddr = (unsigned long)name & PAGE_MASK;
 131         page_offset = (unsigned long)name - page_start_vaddr;
 132         num_pages = DIV_ROUND_UP(page_offset + max_len, PAGE_SIZE);
 133
 134         seq_puts(m, "[anon:");
 135
 136         for (i = 0; i < num_pages; i++) {
 137                 int len;
 138                 int write_len;
 139                 const char *kaddr;
 140                 long pages_pinned;
 141                 struct page *page;
 142
 143                 pages_pinned = get_user_pages(current, mm, page_start_vaddr,
 144                                 1, 0, 0, &page, NULL);
 145                 if (pages_pinned < 1) {
 146                         seq_puts(m, "<fault>]");
 147                         return;
 148                 }
 149
 150                 kaddr = (const char *)kmap(page);
 151                 len = min(max_len, PAGE_SIZE - page_offset);
 152                 write_len = strnlen(kaddr + page_offset, len);
 153                 seq_write(m, kaddr + page_offset, write_len);
 154                 kunmap(page);
 155                 put_page(page);
 156
 157                 /* if strnlen hit a null terminator then we're done */
 158                 if (write_len != len)
 159                         break;
 160
 161                 max_len -= len;
 162                 page_offset = 0;
 163                 page_start_vaddr += PAGE_SIZE;
 164         }
 165
 166         seq_putc(m, ']');
 167 }
 168
 169 static void vma_stop(struct proc_maps_private *priv)
 170 {
 171         struct mm_struct *mm = priv->mm;
 172
 173         release_task_mempolicy(priv);
 174         up_read(&mm->mmap_sem);
 175         mmput(mm);
 176 }
 177
 178 static struct vm_area_struct *
 179 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
 180 {
 181         if (vma == priv->tail_vma)
 182                 return NULL;
 183         return vma->vm_next ?: priv->tail_vma;
 184 }
 185
 186 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
 187 {
 188         if (m->count < m->size) /* vma is copied successfully */
 189                 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
 190 }
 191
 192 static void *m_start(struct seq_file *m, loff_t *ppos)
 193 {
 194         struct proc_maps_private *priv = m->private;
 195         unsigned long last_addr = m->version;
 196         struct mm_struct *mm;
 197         struct vm_area_struct *vma;
 198         unsigned int pos = *ppos;
 199
 200         /* See m_cache_vma(). Zero at the start or after lseek. */
 201         if (last_addr == -1UL)
 202                 return NULL;
 203
 204         priv->task = get_proc_task(priv->inode);
 205         if (!priv->task)
 206                 return ERR_PTR(-ESRCH);
 207
 208         mm = priv->mm;
 209         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
 210                 return NULL;
 211
 212         down_read(&mm->mmap_sem);
 213         hold_task_mempolicy(priv);
 214         priv->tail_vma = get_gate_vma(mm);
 215
 216         if (last_addr) {
 217                 vma = find_vma(mm, last_addr);
 218                 if (vma && (vma = m_next_vma(priv, vma)))
 219                         return vma;
 220         }
 221
 222         m->version = 0;
 223         if (pos < mm->map_count) {
 224                 for (vma = mm->mmap; pos; pos--) {
 225                         m->version = vma->vm_start;
 226                         vma = vma->vm_next;
 227                 }
 228                 return vma;
 229         }
 230
 231         /* we do not bother to update m->version in this case */
 232         if (pos == mm->map_count && priv->tail_vma)
 233                 return priv->tail_vma;
 234
 235         vma_stop(priv);
 236         return NULL;
 237 }
 238
 239 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 240 {
 241         struct proc_maps_private *priv = m->private;
 242         struct vm_area_struct *next;
 243
 244         (*pos)++;
 245         next = m_next_vma(priv, v);
 246         if (!next)
 247                 vma_stop(priv);
 248         return next;
 249 }
 250
 251 static void m_stop(struct seq_file *m, void *v)
 252 {
 253         struct proc_maps_private *priv = m->private;
 254
 255         if (!IS_ERR_OR_NULL(v))
 256                 vma_stop(priv);
 257         if (priv->task) {
 258                 put_task_struct(priv->task);
 259                 priv->task = NULL;
 260         }
 261 }
 262
 263 static int proc_maps_open(struct inode *inode, struct file *file,
 264                         const struct seq_operations *ops, int psize)
 265 {
 266         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 267
 268         if (!priv)
 269                 return -ENOMEM;
 270
 271         priv->inode = inode;
 272         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 273         if (IS_ERR(priv->mm)) {
 274                 int err = PTR_ERR(priv->mm);
 275
 276                 seq_release_private(inode, file);
 277                 return err;
 278         }
 279
 280         return 0;
 281 }
 282
 283 static int proc_map_release(struct inode *inode, struct file *file)
 284 {
 285         struct seq_file *seq = file->private_data;
 286         struct proc_maps_private *priv = seq->private;
 287
 288         if (priv->mm)
 289                 mmdrop(priv->mm);
 290
 291         return seq_release_private(inode, file);
 292 }
 293
 294 static int do_maps_open(struct inode *inode, struct file *file,
 295                         const struct seq_operations *ops)
 296 {
 297         return proc_maps_open(inode, file, ops,
 298                                 sizeof(struct proc_maps_private));
 299 }
 300
 301 static pid_t pid_of_stack(struct proc_maps_private *priv,
 302                                 struct vm_area_struct *vma, bool is_pid)
 303 {
 304         struct inode *inode = priv->inode;
 305         struct task_struct *task;
 306         pid_t ret = 0;
 307
 308         rcu_read_lock();
 309         task = pid_task(proc_pid(inode), PIDTYPE_PID);
 310         if (task) {
 311                 task = task_of_stack(task, vma, is_pid);
 312                 if (task)
 313                         ret = task_pid_nr_ns(task, inode->i_sb->s_fs_info);
 314         }
 315         rcu_read_unlock();
 316
 317         return ret;
 318 }
 319
 320 static void
 321 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 322 {
 323         struct mm_struct *mm = vma->vm_mm;
 324         struct file *file = vma->vm_file;
 325         struct proc_maps_private *priv = m->private;
 326         vm_flags_t flags = vma->vm_flags;
 327         unsigned long ino = 0;
 328         unsigned long long pgoff = 0;
 329         unsigned long start, end;
 330         dev_t dev = 0;
 331         const char *name = NULL;
 332
 333         if (file) {
 334                 struct inode *inode = file_inode(vma->vm_file);
 335                 dev = inode->i_sb->s_dev;
 336                 ino = inode->i_ino;
 337                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 338         }
 339
 340         /* We don't show the stack guard page in /proc/maps */
 341         start = vma->vm_start;
 342         if (stack_guard_page_start(vma, start))
 343                 start += PAGE_SIZE;
 344         end = vma->vm_end;
 345         if (stack_guard_page_end(vma, end))
 346                 end -= PAGE_SIZE;
 347
 348         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 349         seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
 350                         start,
 351                         end,
 352                         flags & VM_READ ? 'r' : '-',
 353                         flags & VM_WRITE ? 'w' : '-',
 354                         flags & VM_EXEC ? 'x' : '-',
 355                         flags & VM_MAYSHARE ? 's' : 'p',
 356                         pgoff,
 357                         MAJOR(dev), MINOR(dev), ino);
 358
 359         /*
 360          * Print the dentry name for named mappings, and a
 361          * special [heap] marker for the heap:
 362          */
 363         if (file) {
 364                 seq_pad(m, ' ');
 365                 seq_file_path(m, file, "\n");
 366                 goto done;
 367         }
 368
 369         if (vma->vm_ops && vma->vm_ops->name) {
 370                 name = vma->vm_ops->name(vma);
 371                 if (name)
 372                         goto done;
 373         }
 374
 375         name = arch_vma_name(vma);
 376         if (!name) {
 377                 pid_t tid;
 378
 379                 if (!mm) {
 380                         name = "[vdso]";
 381                         goto done;
 382                 }
 383
 384                 if (vma->vm_start <= mm->brk &&
 385                     vma->vm_end >= mm->start_brk) {
 386                         name = "[heap]";
 387                         goto done;
 388                 }
 389
 390                 tid = pid_of_stack(priv, vma, is_pid);
 391                 if (tid != 0) {
 392                         /*
 393                          * Thread stack in /proc/PID/task/TID/maps or
 394                          * the main process stack.
 395                          */
 396                         if (!is_pid || (vma->vm_start <= mm->start_stack &&
 397                             vma->vm_end >= mm->start_stack)) {
 398                                 name = "[stack]";
 399                         } else {
 400                                 /* Thread stack in /proc/PID/maps */
 401                                 seq_pad(m, ' ');
 402                                 seq_printf(m, "[stack:%d]", tid);
 403                         }
 404                         goto done;
 405                 }
 406
 407                 if (vma_get_anon_name(vma)) {
 408                         seq_pad(m, ' ');
 409                         seq_print_vma_name(m, vma);
 410                 }
 411         }
 412
 413 done:
 414         if (name) {
 415                 seq_pad(m, ' ');
 416                 seq_puts(m, name);
 417         }
 418         seq_putc(m, '\n');
 419 }
 420
 421 static int show_map(struct seq_file *m, void *v, int is_pid)
 422 {
 423         show_map_vma(m, v, is_pid);
 424         m_cache_vma(m, v);
 425         return 0;
 426 }
 427
 428 static int show_pid_map(struct seq_file *m, void *v)
 429 {
 430         return show_map(m, v, 1);
 431 }
 432
 433 static int show_tid_map(struct seq_file *m, void *v)
 434 {
 435         return show_map(m, v, 0);
 436 }
 437
 438 static const struct seq_operations proc_pid_maps_op = {
 439         .start  = m_start,
 440         .next   = m_next,
 441         .stop   = m_stop,
 442         .show   = show_pid_map
 443 };
 444
 445 static const struct seq_operations proc_tid_maps_op = {
 446         .start  = m_start,
 447         .next   = m_next,
 448         .stop   = m_stop,
 449         .show   = show_tid_map
 450 };
 451
 452 static int pid_maps_open(struct inode *inode, struct file *file)
 453 {
 454         return do_maps_open(inode, file, &proc_pid_maps_op);
 455 }
 456
 457 static int tid_maps_open(struct inode *inode, struct file *file)
 458 {
 459         return do_maps_open(inode, file, &proc_tid_maps_op);
 460 }
 461
 462 const struct file_operations proc_pid_maps_operations = {
 463         .open           = pid_maps_open,
 464         .read           = seq_read,
 465         .llseek         = seq_lseek,
 466         .release        = proc_map_release,
 467 };
 468
 469 const struct file_operations proc_tid_maps_operations = {
 470         .open           = tid_maps_open,
 471         .read           = seq_read,
 472         .llseek         = seq_lseek,
 473         .release        = proc_map_release,
 474 };
 475
 476 /*
 477  * Proportional Set Size(PSS): my share of RSS.
 478  *
 479  * PSS of a process is the count of pages it has in memory, where each
 480  * page is divided by the number of processes sharing it.  So if a
 481  * process has 1000 pages all to itself, and 1000 shared with one other
 482  * process, its PSS will be 1500.
 483  *
 484  * To keep (accumulated) division errors low, we adopt a 64bit
 485  * fixed-point pss counter to minimize division errors. So (pss >>
 486  * PSS_SHIFT) would be the real byte count.
 487  *
 488  * A shift of 12 before division means (assuming 4K page size):
 489  *      - 1M 3-user-pages add up to 8KB errors;
 490  *      - supports mapcount up to 2^24, or 16M;
 491  *      - supports PSS up to 2^52 bytes, or 4PB.
 492  */
 493 #define PSS_SHIFT 12
 494
 495 #ifdef CONFIG_PROC_PAGE_MONITOR
 496 struct mem_size_stats {
 497         unsigned long resident;
 498         unsigned long shared_clean;
 499         unsigned long shared_dirty;
 500         unsigned long private_clean;
 501         unsigned long private_dirty;
 502         unsigned long referenced;
 503         unsigned long anonymous;
 504         unsigned long anonymous_thp;
 505         unsigned long swap;
 506         unsigned long shared_hugetlb;
 507         unsigned long private_hugetlb;
 508         u64 pss;
 509         u64 swap_pss;
 510 };
 511
 512 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 513                 unsigned long size, bool young, bool dirty)
 514 {
 515         int mapcount;
 516
 517         if (PageAnon(page))
 518                 mss->anonymous += size;
 519
 520         mss->resident += size;
 521         /* Accumulate the size in pages that have been accessed. */
 522         if (young || page_is_young(page) || PageReferenced(page))
 523                 mss->referenced += size;
 524         mapcount = page_mapcount(page);
 525         if (mapcount >= 2) {
 526                 u64 pss_delta;
 527
 528                 if (dirty || PageDirty(page))
 529                         mss->shared_dirty += size;
 530                 else
 531                         mss->shared_clean += size;
 532                 pss_delta = (u64)size << PSS_SHIFT;
 533                 do_div(pss_delta, mapcount);
 534                 mss->pss += pss_delta;
 535         } else {
 536                 if (dirty || PageDirty(page))
 537                         mss->private_dirty += size;
 538                 else
 539                         mss->private_clean += size;
 540                 mss->pss += (u64)size << PSS_SHIFT;
 541         }
 542 }
 543
 544 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 545                 struct mm_walk *walk)
 546 {
 547         struct mem_size_stats *mss = walk->private;
 548         struct vm_area_struct *vma = walk->vma;
 549         struct page *page = NULL;
 550
 551         if (pte_present(*pte)) {
 552                 page = vm_normal_page(vma, addr, *pte);
 553         } else if (is_swap_pte(*pte)) {
 554                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 555
 556                 if (!non_swap_entry(swpent)) {
 557                         int mapcount;
 558
 559                         mss->swap += PAGE_SIZE;
 560                         mapcount = swp_swapcount(swpent);
 561                         if (mapcount >= 2) {
 562                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 563
 564                                 do_div(pss_delta, mapcount);
 565                                 mss->swap_pss += pss_delta;
 566                         } else {
 567                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 568                         }
 569                 } else if (is_migration_entry(swpent))
 570                         page = migration_entry_to_page(swpent);
 571         }
 572
 573         if (!page)
 574                 return;
 575         smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte));
 576 }
 577
 578 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 579 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 580                 struct mm_walk *walk)
 581 {
 582         struct mem_size_stats *mss = walk->private;
 583         struct vm_area_struct *vma = walk->vma;
 584         struct page *page;
 585
 586         /* FOLL_DUMP will return -EFAULT on huge zero page */
 587         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 588         if (IS_ERR_OR_NULL(page))
 589                 return;
 590         mss->anonymous_thp += HPAGE_PMD_SIZE;
 591         smaps_account(mss, page, HPAGE_PMD_SIZE,
 592                         pmd_young(*pmd), pmd_dirty(*pmd));
 593 }
 594 #else
 595 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 596                 struct mm_walk *walk)
 597 {
 598 }
 599 #endif
 600
 601 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 602                            struct mm_walk *walk)
 603 {
 604         struct vm_area_struct *vma = walk->vma;
 605         pte_t *pte;
 606         spinlock_t *ptl;
 607
 608         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
 609                 smaps_pmd_entry(pmd, addr, walk);
 610                 spin_unlock(ptl);
 611                 return 0;
 612         }
 613
 614         if (pmd_trans_unstable(pmd))
 615                 return 0;
 616         /*
 617          * The mmap_sem held all the way back in m_start() is what
 618          * keeps khugepaged out of here and from collapsing things
 619          * in here.
 620          */
 621         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 622         for (; addr != end; pte++, addr += PAGE_SIZE)
 623                 smaps_pte_entry(pte, addr, walk);
 624         pte_unmap_unlock(pte - 1, ptl);
 625         cond_resched();
 626         return 0;
 627 }
 628
 629 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 630 {
 631         /*
 632          * Don't forget to update Documentation/ on changes.
 633          */
 634         static const char mnemonics[BITS_PER_LONG][2] = {
 635                 /*
 636                  * In case if we meet a flag we don't know about.
 637                  */
 638                 [0 ... (BITS_PER_LONG-1)] = "??",
 639
 640                 [ilog2(VM_READ)]        = "rd",
 641                 [ilog2(VM_WRITE)]       = "wr",
 642                 [ilog2(VM_EXEC)]        = "ex",
 643                 [ilog2(VM_SHARED)]      = "sh",
 644                 [ilog2(VM_MAYREAD)]     = "mr",
 645                 [ilog2(VM_MAYWRITE)]    = "mw",
 646                 [ilog2(VM_MAYEXEC)]     = "me",
 647                 [ilog2(VM_MAYSHARE)]    = "ms",
 648                 [ilog2(VM_GROWSDOWN)]   = "gd",
 649                 [ilog2(VM_PFNMAP)]      = "pf",
 650                 [ilog2(VM_DENYWRITE)]   = "dw",
 651 #ifdef CONFIG_X86_INTEL_MPX
 652                 [ilog2(VM_MPX)]         = "mp",
 653 #endif
 654                 [ilog2(VM_LOCKED)]      = "lo",
 655                 [ilog2(VM_IO)]          = "io",
 656                 [ilog2(VM_SEQ_READ)]    = "sr",
 657                 [ilog2(VM_RAND_READ)]   = "rr",
 658                 [ilog2(VM_DONTCOPY)]    = "dc",
 659                 [ilog2(VM_DONTEXPAND)]  = "de",
 660                 [ilog2(VM_ACCOUNT)]     = "ac",
 661                 [ilog2(VM_NORESERVE)]   = "nr",
 662                 [ilog2(VM_HUGETLB)]     = "ht",
 663                 [ilog2(VM_ARCH_1)]      = "ar",
 664                 [ilog2(VM_DONTDUMP)]    = "dd",
 665 #ifdef CONFIG_MEM_SOFT_DIRTY
 666                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 667 #endif
 668                 [ilog2(VM_MIXEDMAP)]    = "mm",
 669                 [ilog2(VM_HUGEPAGE)]    = "hg",
 670                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 671                 [ilog2(VM_MERGEABLE)]   = "mg",
 672                 [ilog2(VM_UFFD_MISSING)]= "um",
 673                 [ilog2(VM_UFFD_WP)]     = "uw",
 674         };
 675         size_t i;
 676
 677         seq_puts(m, "VmFlags: ");
 678         for (i = 0; i < BITS_PER_LONG; i++) {
 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 static int show_smap(struct seq_file *m, void *v, int is_pid)
 717 {
 718         struct vm_area_struct *vma = v;
 719         struct mem_size_stats mss;
 720         struct mm_walk smaps_walk = {
 721                 .pmd_entry = smaps_pte_range,
 722 #ifdef CONFIG_HUGETLB_PAGE
 723                 .hugetlb_entry = smaps_hugetlb_range,
 724 #endif
 725                 .mm = vma->vm_mm,
 726                 .private = &mss,
 727         };
 728
 729         memset(&mss, 0, sizeof mss);
 730         /* mmap_sem is held in m_start */
 731         walk_page_vma(vma, &smaps_walk);
 732
 733         show_map_vma(m, vma, is_pid);
 734
 735         if (vma_get_anon_name(vma)) {
 736                 seq_puts(m, "Name:           ");
 737                 seq_print_vma_name(m, vma);
 738                 seq_putc(m, '\n');
 739         }
 740
 741         seq_printf(m,
 742                    "Size:           %8lu kB\n"
 743                    "Rss:            %8lu kB\n"
 744                    "Pss:            %8lu kB\n"
 745                    "Shared_Clean:   %8lu kB\n"
 746                    "Shared_Dirty:   %8lu kB\n"
 747                    "Private_Clean:  %8lu kB\n"
 748                    "Private_Dirty:  %8lu kB\n"
 749                    "Referenced:     %8lu kB\n"
 750                    "Anonymous:      %8lu kB\n"
 751                    "AnonHugePages:  %8lu kB\n"
 752                    "Shared_Hugetlb: %8lu kB\n"
 753                    "Private_Hugetlb: %7lu kB\n"
 754                    "Swap:           %8lu kB\n"
 755                    "SwapPss:        %8lu kB\n"
 756                    "KernelPageSize: %8lu kB\n"
 757                    "MMUPageSize:    %8lu kB\n"
 758                    "Locked:         %8lu kB\n",
 759                    (vma->vm_end - vma->vm_start) >> 10,
 760                    mss.resident >> 10,
 761                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 762                    mss.shared_clean  >> 10,
 763                    mss.shared_dirty  >> 10,
 764                    mss.private_clean >> 10,
 765                    mss.private_dirty >> 10,
 766                    mss.referenced >> 10,
 767                    mss.anonymous >> 10,
 768                    mss.anonymous_thp >> 10,
 769                    mss.shared_hugetlb >> 10,
 770                    mss.private_hugetlb >> 10,
 771                    mss.swap >> 10,
 772                    (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
 773                    vma_kernel_pagesize(vma) >> 10,
 774                    vma_mmu_pagesize(vma) >> 10,
 775                    (vma->vm_flags & VM_LOCKED) ?
 776                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 777
 778         show_smap_vma_flags(m, vma);
 779         m_cache_vma(m, vma);
 780         return 0;
 781 }
 782
 783 static int show_pid_smap(struct seq_file *m, void *v)
 784 {
 785         return show_smap(m, v, 1);
 786 }
 787
 788 static int show_tid_smap(struct seq_file *m, void *v)
 789 {
 790         return show_smap(m, v, 0);
 791 }
 792
 793 static const struct seq_operations proc_pid_smaps_op = {
 794         .start  = m_start,
 795         .next   = m_next,
 796         .stop   = m_stop,
 797         .show   = show_pid_smap
 798 };
 799
 800 static const struct seq_operations proc_tid_smaps_op = {
 801         .start  = m_start,
 802         .next   = m_next,
 803         .stop   = m_stop,
 804         .show   = show_tid_smap
 805 };
 806
 807 static int pid_smaps_open(struct inode *inode, struct file *file)
 808 {
 809         return do_maps_open(inode, file, &proc_pid_smaps_op);
 810 }
 811
 812 static int tid_smaps_open(struct inode *inode, struct file *file)
 813 {
 814         return do_maps_open(inode, file, &proc_tid_smaps_op);
 815 }
 816
 817 const struct file_operations proc_pid_smaps_operations = {
 818         .open           = pid_smaps_open,
 819         .read           = seq_read,
 820         .llseek         = seq_lseek,
 821         .release        = proc_map_release,
 822 };
 823
 824 const struct file_operations proc_tid_smaps_operations = {
 825         .open           = tid_smaps_open,
 826         .read           = seq_read,
 827         .llseek         = seq_lseek,
 828         .release        = proc_map_release,
 829 };
 830
 831 enum clear_refs_types {
 832         CLEAR_REFS_ALL = 1,
 833         CLEAR_REFS_ANON,
 834         CLEAR_REFS_MAPPED,
 835         CLEAR_REFS_SOFT_DIRTY,
 836         CLEAR_REFS_MM_HIWATER_RSS,
 837         CLEAR_REFS_LAST,
 838 };
 839
 840 struct clear_refs_private {
 841         enum clear_refs_types type;
 842 };
 843
 844 #ifdef CONFIG_MEM_SOFT_DIRTY
 845 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 846                 unsigned long addr, pte_t *pte)
 847 {
 848         /*
 849          * The soft-dirty tracker uses #PF-s to catch writes
 850          * to pages, so write-protect the pte as well. See the
 851          * Documentation/vm/soft-dirty.txt for full description
 852          * of how soft-dirty works.
 853          */
 854         pte_t ptent = *pte;
 855
 856         if (pte_present(ptent)) {
 857                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 858                 ptent = pte_wrprotect(ptent);
 859                 ptent = pte_clear_soft_dirty(ptent);
 860                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 861         } else if (is_swap_pte(ptent)) {
 862                 ptent = pte_swp_clear_soft_dirty(ptent);
 863                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 864         }
 865 }
 866 #else
 867 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 868                 unsigned long addr, pte_t *pte)
 869 {
 870 }
 871 #endif
 872
 873 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 874 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 875                 unsigned long addr, pmd_t *pmdp)
 876 {
 877         pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
 878
 879         pmd = pmd_wrprotect(pmd);
 880         pmd = pmd_clear_soft_dirty(pmd);
 881
 882         if (vma->vm_flags & VM_SOFTDIRTY)
 883                 vma->vm_flags &= ~VM_SOFTDIRTY;
 884
 885         set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 886 }
 887 #else
 888 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 889                 unsigned long addr, pmd_t *pmdp)
 890 {
 891 }
 892 #endif
 893
 894 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 895                                 unsigned long end, struct mm_walk *walk)
 896 {
 897         struct clear_refs_private *cp = walk->private;
 898         struct vm_area_struct *vma = walk->vma;
 899         pte_t *pte, ptent;
 900         spinlock_t *ptl;
 901         struct page *page;
 902
 903         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
 904                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 905                         clear_soft_dirty_pmd(vma, addr, pmd);
 906                         goto out;
 907                 }
 908
 909                 page = pmd_page(*pmd);
 910
 911                 /* Clear accessed and referenced bits. */
 912                 pmdp_test_and_clear_young(vma, addr, pmd);
 913                 test_and_clear_page_young(page);
 914                 ClearPageReferenced(page);
 915 out:
 916                 spin_unlock(ptl);
 917                 return 0;
 918         }
 919
 920         if (pmd_trans_unstable(pmd))
 921                 return 0;
 922
 923         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 924         for (; addr != end; pte++, addr += PAGE_SIZE) {
 925                 ptent = *pte;
 926
 927                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 928                         clear_soft_dirty(vma, addr, pte);
 929                         continue;
 930                 }
 931
 932                 if (!pte_present(ptent))
 933                         continue;
 934
 935                 page = vm_normal_page(vma, addr, ptent);
 936                 if (!page)
 937                         continue;
 938
 939                 /* Clear accessed and referenced bits. */
 940                 ptep_test_and_clear_young(vma, addr, pte);
 941                 test_and_clear_page_young(page);
 942                 ClearPageReferenced(page);
 943         }
 944         pte_unmap_unlock(pte - 1, ptl);
 945         cond_resched();
 946         return 0;
 947 }
 948
 949 static int clear_refs_test_walk(unsigned long start, unsigned long end,
 950                                 struct mm_walk *walk)
 951 {
 952         struct clear_refs_private *cp = walk->private;
 953         struct vm_area_struct *vma = walk->vma;
 954
 955         if (vma->vm_flags & VM_PFNMAP)
 956                 return 1;
 957
 958         /*
 959          * Writing 1 to /proc/pid/clear_refs affects all pages.
 960          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
 961          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
 962          * Writing 4 to /proc/pid/clear_refs affects all pages.
 963          */
 964         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
 965                 return 1;
 966         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
 967                 return 1;
 968         return 0;
 969 }
 970
 971 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 972                                 size_t count, loff_t *ppos)
 973 {
 974         struct task_struct *task;
 975         char buffer[PROC_NUMBUF];
 976         struct mm_struct *mm;
 977         struct vm_area_struct *vma;
 978         enum clear_refs_types type;
 979         int itype;
 980         int rv;
 981
 982         memset(buffer, 0, sizeof(buffer));
 983         if (count > sizeof(buffer) - 1)
 984                 count = sizeof(buffer) - 1;
 985         if (copy_from_user(buffer, buf, count))
 986                 return -EFAULT;
 987         rv = kstrtoint(strstrip(buffer), 10, &itype);
 988         if (rv < 0)
 989                 return rv;
 990         type = (enum clear_refs_types)itype;
 991         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
 992                 return -EINVAL;
 993
 994         task = get_proc_task(file_inode(file));
 995         if (!task)
 996                 return -ESRCH;
 997         mm = get_task_mm(task);
 998         if (mm) {
 999                 struct clear_refs_private cp = {
1000                         .type = type,
1001                 };
1002                 struct mm_walk clear_refs_walk = {
1003                         .pmd_entry = clear_refs_pte_range,
1004                         .test_walk = clear_refs_test_walk,
1005                         .mm = mm,
1006                         .private = &cp,
1007                 };
1008
1009                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1010                         /*
1011                          * Writing 5 to /proc/pid/clear_refs resets the peak
1012                          * resident set size to this mm's current rss value.
1013                          */
1014                         down_write(&mm->mmap_sem);
1015                         reset_mm_hiwater_rss(mm);
1016                         up_write(&mm->mmap_sem);
1017                         goto out_mm;
1018                 }
1019
1020                 down_read(&mm->mmap_sem);
1021                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1022                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1023                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1024                                         continue;
1025                                 up_read(&mm->mmap_sem);
1026                                 down_write(&mm->mmap_sem);
1027                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1028                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1029                                         vma_set_page_prot(vma);
1030                                 }
1031                                 downgrade_write(&mm->mmap_sem);
1032                                 break;
1033                         }
1034                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1035                 }
1036                 walk_page_range(0, ~0UL, &clear_refs_walk);
1037                 if (type == CLEAR_REFS_SOFT_DIRTY)
1038                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1039                 flush_tlb_mm(mm);
1040                 up_read(&mm->mmap_sem);
1041 out_mm:
1042                 mmput(mm);
1043         }
1044         put_task_struct(task);
1045
1046         return count;
1047 }
1048
1049 const struct file_operations proc_clear_refs_operations = {
1050         .write          = clear_refs_write,
1051         .llseek         = noop_llseek,
1052 };
1053
1054 typedef struct {
1055         u64 pme;
1056 } pagemap_entry_t;
1057
1058 struct pagemapread {
1059         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1060         pagemap_entry_t *buffer;
1061         bool show_pfn;
1062 };
1063
1064 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1065 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1066
1067 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1068 #define PM_PFRAME_BITS          55
1069 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1070 #define PM_SOFT_DIRTY           BIT_ULL(55)
1071 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1072 #define PM_FILE                 BIT_ULL(61)
1073 #define PM_SWAP                 BIT_ULL(62)
1074 #define PM_PRESENT              BIT_ULL(63)
1075
1076 #define PM_END_OF_BUFFER    1
1077
1078 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1079 {
1080         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1081 }
1082
1083 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1084                           struct pagemapread *pm)
1085 {
1086         pm->buffer[pm->pos++] = *pme;
1087         if (pm->pos >= pm->len)
1088                 return PM_END_OF_BUFFER;
1089         return 0;
1090 }
1091
1092 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1093                                 struct mm_walk *walk)
1094 {
1095         struct pagemapread *pm = walk->private;
1096         unsigned long addr = start;
1097         int err = 0;
1098
1099         while (addr < end) {
1100                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1101                 pagemap_entry_t pme = make_pme(0, 0);
1102                 /* End of address space hole, which we mark as non-present. */
1103                 unsigned long hole_end;
1104
1105                 if (vma)
1106                         hole_end = min(end, vma->vm_start);
1107                 else
1108                         hole_end = end;
1109
1110                 for (; addr < hole_end; addr += PAGE_SIZE) {
1111                         err = add_to_pagemap(addr, &pme, pm);
1112                         if (err)
1113                                 goto out;
1114                 }
1115
1116                 if (!vma)
1117                         break;
1118
1119                 /* Addresses in the VMA. */
1120                 if (vma->vm_flags & VM_SOFTDIRTY)
1121                         pme = make_pme(0, PM_SOFT_DIRTY);
1122                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1123                         err = add_to_pagemap(addr, &pme, pm);
1124                         if (err)
1125                                 goto out;
1126                 }
1127         }
1128 out:
1129         return err;
1130 }
1131
1132 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1133                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1134 {
1135         u64 frame = 0, flags = 0;
1136         struct page *page = NULL;
1137
1138         if (pte_present(pte)) {
1139                 if (pm->show_pfn)
1140                         frame = pte_pfn(pte);
1141                 flags |= PM_PRESENT;
1142                 page = vm_normal_page(vma, addr, pte);
1143                 if (pte_soft_dirty(pte))
1144                         flags |= PM_SOFT_DIRTY;
1145         } else if (is_swap_pte(pte)) {
1146                 swp_entry_t entry;
1147                 if (pte_swp_soft_dirty(pte))
1148                         flags |= PM_SOFT_DIRTY;
1149                 entry = pte_to_swp_entry(pte);
1150                 frame = swp_type(entry) |
1151                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1152                 flags |= PM_SWAP;
1153                 if (is_migration_entry(entry))
1154                         page = migration_entry_to_page(entry);
1155         }
1156
1157         if (page && !PageAnon(page))
1158                 flags |= PM_FILE;
1159         if (page && page_mapcount(page) == 1)
1160                 flags |= PM_MMAP_EXCLUSIVE;
1161         if (vma->vm_flags & VM_SOFTDIRTY)
1162                 flags |= PM_SOFT_DIRTY;
1163
1164         return make_pme(frame, flags);
1165 }
1166
1167 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1168                              struct mm_walk *walk)
1169 {
1170         struct vm_area_struct *vma = walk->vma;
1171         struct pagemapread *pm = walk->private;
1172         spinlock_t *ptl;
1173         pte_t *pte, *orig_pte;
1174         int err = 0;
1175
1176 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1177         if (pmd_trans_huge_lock(pmdp, vma, &ptl) == 1) {
1178                 u64 flags = 0, frame = 0;
1179                 pmd_t pmd = *pmdp;
1180
1181                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1182                         flags |= PM_SOFT_DIRTY;
1183
1184                 /*
1185                  * Currently pmd for thp is always present because thp
1186                  * can not be swapped-out, migrated, or HWPOISONed
1187                  * (split in such cases instead.)
1188                  * This if-check is just to prepare for future implementation.
1189                  */
1190                 if (pmd_present(pmd)) {
1191                         struct page *page = pmd_page(pmd);
1192
1193                         if (page_mapcount(page) == 1)
1194                                 flags |= PM_MMAP_EXCLUSIVE;
1195
1196                         flags |= PM_PRESENT;
1197                         if (pm->show_pfn)
1198                                 frame = pmd_pfn(pmd) +
1199                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1200                 }
1201
1202                 for (; addr != end; addr += PAGE_SIZE) {
1203                         pagemap_entry_t pme = make_pme(frame, flags);
1204
1205                         err = add_to_pagemap(addr, &pme, pm);
1206                         if (err)
1207                                 break;
1208                         if (pm->show_pfn && (flags & PM_PRESENT))
1209                                 frame++;
1210                 }
1211                 spin_unlock(ptl);
1212                 return err;
1213         }
1214
1215         if (pmd_trans_unstable(pmdp))
1216                 return 0;
1217 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1218
1219         /*
1220          * We can assume that @vma always points to a valid one and @end never
1221          * goes beyond vma->vm_end.
1222          */
1223         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1224         for (; addr < end; pte++, addr += PAGE_SIZE) {
1225                 pagemap_entry_t pme;
1226
1227                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1228                 err = add_to_pagemap(addr, &pme, pm);
1229                 if (err)
1230                         break;
1231         }
1232         pte_unmap_unlock(orig_pte, ptl);
1233
1234         cond_resched();
1235
1236         return err;
1237 }
1238
1239 #ifdef CONFIG_HUGETLB_PAGE
1240 /* This function walks within one hugetlb entry in the single call */
1241 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1242                                  unsigned long addr, unsigned long end,
1243                                  struct mm_walk *walk)
1244 {
1245         struct pagemapread *pm = walk->private;
1246         struct vm_area_struct *vma = walk->vma;
1247         u64 flags = 0, frame = 0;
1248         int err = 0;
1249         pte_t pte;
1250
1251         if (vma->vm_flags & VM_SOFTDIRTY)
1252                 flags |= PM_SOFT_DIRTY;
1253
1254         pte = huge_ptep_get(ptep);
1255         if (pte_present(pte)) {
1256                 struct page *page = pte_page(pte);
1257
1258                 if (!PageAnon(page))
1259                         flags |= PM_FILE;
1260
1261                 if (page_mapcount(page) == 1)
1262                         flags |= PM_MMAP_EXCLUSIVE;
1263
1264                 flags |= PM_PRESENT;
1265                 if (pm->show_pfn)
1266                         frame = pte_pfn(pte) +
1267                                 ((addr & ~hmask) >> PAGE_SHIFT);
1268         }
1269
1270         for (; addr != end; addr += PAGE_SIZE) {
1271                 pagemap_entry_t pme = make_pme(frame, flags);
1272
1273                 err = add_to_pagemap(addr, &pme, pm);
1274                 if (err)
1275                         return err;
1276                 if (pm->show_pfn && (flags & PM_PRESENT))
1277                         frame++;
1278         }
1279
1280         cond_resched();
1281
1282         return err;
1283 }
1284 #endif /* HUGETLB_PAGE */
1285
1286 /*
1287  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1288  *
1289  * For each page in the address space, this file contains one 64-bit entry
1290  * consisting of the following:
1291  *
1292  * Bits 0-54  page frame number (PFN) if present
1293  * Bits 0-4   swap type if swapped
1294  * Bits 5-54  swap offset if swapped
1295  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1296  * Bit  56    page exclusively mapped
1297  * Bits 57-60 zero
1298  * Bit  61    page is file-page or shared-anon
1299  * Bit  62    page swapped
1300  * Bit  63    page present
1301  *
1302  * If the page is not present but in swap, then the PFN contains an
1303  * encoding of the swap file number and the page's offset into the
1304  * swap. Unmapped pages return a null PFN. This allows determining
1305  * precisely which pages are mapped (or in swap) and comparing mapped
1306  * pages between processes.
1307  *
1308  * Efficient users of this interface will use /proc/pid/maps to
1309  * determine which areas of memory are actually mapped and llseek to
1310  * skip over unmapped regions.
1311  */
1312 static ssize_t pagemap_read(struct file *file, char __user *buf,
1313                             size_t count, loff_t *ppos)
1314 {
1315         struct mm_struct *mm = file->private_data;
1316         struct pagemapread pm;
1317         struct mm_walk pagemap_walk = {};
1318         unsigned long src;
1319         unsigned long svpfn;
1320         unsigned long start_vaddr;
1321         unsigned long end_vaddr;
1322         int ret = 0, copied = 0;
1323
1324         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1325                 goto out;
1326
1327         ret = -EINVAL;
1328         /* file position must be aligned */
1329         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1330                 goto out_mm;
1331
1332         ret = 0;
1333         if (!count)
1334                 goto out_mm;
1335
1336         /* do not disclose physical addresses: attack vector */
1337         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1338
1339         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1340         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1341         ret = -ENOMEM;
1342         if (!pm.buffer)
1343                 goto out_mm;
1344
1345         pagemap_walk.pmd_entry = pagemap_pmd_range;
1346         pagemap_walk.pte_hole = pagemap_pte_hole;
1347 #ifdef CONFIG_HUGETLB_PAGE
1348         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1349 #endif
1350         pagemap_walk.mm = mm;
1351         pagemap_walk.private = &pm;
1352
1353         src = *ppos;
1354         svpfn = src / PM_ENTRY_BYTES;
1355         start_vaddr = svpfn << PAGE_SHIFT;
1356         end_vaddr = mm->task_size;
1357
1358         /* watch out for wraparound */
1359         if (svpfn > mm->task_size >> PAGE_SHIFT)
1360                 start_vaddr = end_vaddr;
1361
1362         /*
1363          * The odds are that this will stop walking way
1364          * before end_vaddr, because the length of the
1365          * user buffer is tracked in "pm", and the walk
1366          * will stop when we hit the end of the buffer.
1367          */
1368         ret = 0;
1369         while (count && (start_vaddr < end_vaddr)) {
1370                 int len;
1371                 unsigned long end;
1372
1373                 pm.pos = 0;
1374                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1375                 /* overflow ? */
1376                 if (end < start_vaddr || end > end_vaddr)
1377                         end = end_vaddr;
1378                 down_read(&mm->mmap_sem);
1379                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1380                 up_read(&mm->mmap_sem);
1381                 start_vaddr = end;
1382
1383                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1384                 if (copy_to_user(buf, pm.buffer, len)) {
1385                         ret = -EFAULT;
1386                         goto out_free;
1387                 }
1388                 copied += len;
1389                 buf += len;
1390                 count -= len;
1391         }
1392         *ppos += copied;
1393         if (!ret || ret == PM_END_OF_BUFFER)
1394                 ret = copied;
1395
1396 out_free:
1397         kfree(pm.buffer);
1398 out_mm:
1399         mmput(mm);
1400 out:
1401         return ret;
1402 }
1403
1404 static int pagemap_open(struct inode *inode, struct file *file)
1405 {
1406         struct mm_struct *mm;
1407
1408         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1409         if (IS_ERR(mm))
1410                 return PTR_ERR(mm);
1411         file->private_data = mm;
1412         return 0;
1413 }
1414
1415 static int pagemap_release(struct inode *inode, struct file *file)
1416 {
1417         struct mm_struct *mm = file->private_data;
1418
1419         if (mm)
1420                 mmdrop(mm);
1421         return 0;
1422 }
1423
1424 const struct file_operations proc_pagemap_operations = {
1425         .llseek         = mem_lseek, /* borrow this */
1426         .read           = pagemap_read,
1427         .open           = pagemap_open,
1428         .release        = pagemap_release,
1429 };
1430 #endif /* CONFIG_PROC_PAGE_MONITOR */
1431
1432 #ifdef CONFIG_NUMA
1433
1434 struct numa_maps {
1435         unsigned long pages;
1436         unsigned long anon;
1437         unsigned long active;
1438         unsigned long writeback;
1439         unsigned long mapcount_max;
1440         unsigned long dirty;
1441         unsigned long swapcache;
1442         unsigned long node[MAX_NUMNODES];
1443 };
1444
1445 struct numa_maps_private {
1446         struct proc_maps_private proc_maps;
1447         struct numa_maps md;
1448 };
1449
1450 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1451                         unsigned long nr_pages)
1452 {
1453         int count = page_mapcount(page);
1454
1455         md->pages += nr_pages;
1456         if (pte_dirty || PageDirty(page))
1457                 md->dirty += nr_pages;
1458
1459         if (PageSwapCache(page))
1460                 md->swapcache += nr_pages;
1461
1462         if (PageActive(page) || PageUnevictable(page))
1463                 md->active += nr_pages;
1464
1465         if (PageWriteback(page))
1466                 md->writeback += nr_pages;
1467
1468         if (PageAnon(page))
1469                 md->anon += nr_pages;
1470
1471         if (count > md->mapcount_max)
1472                 md->mapcount_max = count;
1473
1474         md->node[page_to_nid(page)] += nr_pages;
1475 }
1476
1477 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1478                 unsigned long addr)
1479 {
1480         struct page *page;
1481         int nid;
1482
1483         if (!pte_present(pte))
1484                 return NULL;
1485
1486         page = vm_normal_page(vma, addr, pte);
1487         if (!page)
1488                 return NULL;
1489
1490         if (PageReserved(page))
1491                 return NULL;
1492
1493         nid = page_to_nid(page);
1494         if (!node_isset(nid, node_states[N_MEMORY]))
1495                 return NULL;
1496
1497         return page;
1498 }
1499
1500 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1501 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1502                                               struct vm_area_struct *vma,
1503                                               unsigned long addr)
1504 {
1505         struct page *page;
1506         int nid;
1507
1508         if (!pmd_present(pmd))
1509                 return NULL;
1510
1511         page = vm_normal_page_pmd(vma, addr, pmd);
1512         if (!page)
1513                 return NULL;
1514
1515         if (PageReserved(page))
1516                 return NULL;
1517
1518         nid = page_to_nid(page);
1519         if (!node_isset(nid, node_states[N_MEMORY]))
1520                 return NULL;
1521
1522         return page;
1523 }
1524 #endif
1525
1526 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1527                 unsigned long end, struct mm_walk *walk)
1528 {
1529         struct numa_maps *md = walk->private;
1530         struct vm_area_struct *vma = walk->vma;
1531         spinlock_t *ptl;
1532         pte_t *orig_pte;
1533         pte_t *pte;
1534
1535 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1536         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1537                 struct page *page;
1538
1539                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1540                 if (page)
1541                         gather_stats(page, md, pmd_dirty(*pmd),
1542                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1543                 spin_unlock(ptl);
1544                 return 0;
1545         }
1546
1547         if (pmd_trans_unstable(pmd))
1548                 return 0;
1549 #endif
1550         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1551         do {
1552                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1553                 if (!page)
1554                         continue;
1555                 gather_stats(page, md, pte_dirty(*pte), 1);
1556
1557         } while (pte++, addr += PAGE_SIZE, addr != end);
1558         pte_unmap_unlock(orig_pte, ptl);
1559         return 0;
1560 }
1561 #ifdef CONFIG_HUGETLB_PAGE
1562 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1563                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1564 {
1565         pte_t huge_pte = huge_ptep_get(pte);
1566         struct numa_maps *md;
1567         struct page *page;
1568
1569         if (!pte_present(huge_pte))
1570                 return 0;
1571
1572         page = pte_page(huge_pte);
1573         if (!page)
1574                 return 0;
1575
1576         md = walk->private;
1577         gather_stats(page, md, pte_dirty(huge_pte), 1);
1578         return 0;
1579 }
1580
1581 #else
1582 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1583                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1584 {
1585         return 0;
1586 }
1587 #endif
1588
1589 /*
1590  * Display pages allocated per node and memory policy via /proc.
1591  */
1592 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1593 {
1594         struct numa_maps_private *numa_priv = m->private;
1595         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1596         struct vm_area_struct *vma = v;
1597         struct numa_maps *md = &numa_priv->md;
1598         struct file *file = vma->vm_file;
1599         struct mm_struct *mm = vma->vm_mm;
1600         struct mm_walk walk = {
1601                 .hugetlb_entry = gather_hugetlb_stats,
1602                 .pmd_entry = gather_pte_stats,
1603                 .private = md,
1604                 .mm = mm,
1605         };
1606         struct mempolicy *pol;
1607         char buffer[64];
1608         int nid;
1609
1610         if (!mm)
1611                 return 0;
1612
1613         /* Ensure we start with an empty set of numa_maps statistics. */
1614         memset(md, 0, sizeof(*md));
1615
1616         pol = __get_vma_policy(vma, vma->vm_start);
1617         if (pol) {
1618                 mpol_to_str(buffer, sizeof(buffer), pol);
1619                 mpol_cond_put(pol);
1620         } else {
1621                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1622         }
1623
1624         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1625
1626         if (file) {
1627                 seq_puts(m, " file=");
1628                 seq_file_path(m, file, "\n\t= ");
1629         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1630                 seq_puts(m, " heap");
1631         } else {
1632                 pid_t tid = pid_of_stack(proc_priv, vma, is_pid);
1633                 if (tid != 0) {
1634                         /*
1635                          * Thread stack in /proc/PID/task/TID/maps or
1636                          * the main process stack.
1637                          */
1638                         if (!is_pid || (vma->vm_start <= mm->start_stack &&
1639                             vma->vm_end >= mm->start_stack))
1640                                 seq_puts(m, " stack");
1641                         else
1642                                 seq_printf(m, " stack:%d", tid);
1643                 }
1644         }
1645
1646         if (is_vm_hugetlb_page(vma))
1647                 seq_puts(m, " huge");
1648
1649         /* mmap_sem is held by m_start */
1650         walk_page_vma(vma, &walk);
1651
1652         if (!md->pages)
1653                 goto out;
1654
1655         if (md->anon)
1656                 seq_printf(m, " anon=%lu", md->anon);
1657
1658         if (md->dirty)
1659                 seq_printf(m, " dirty=%lu", md->dirty);
1660
1661         if (md->pages != md->anon && md->pages != md->dirty)
1662                 seq_printf(m, " mapped=%lu", md->pages);
1663
1664         if (md->mapcount_max > 1)
1665                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1666
1667         if (md->swapcache)
1668                 seq_printf(m, " swapcache=%lu", md->swapcache);
1669
1670         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1671                 seq_printf(m, " active=%lu", md->active);
1672
1673         if (md->writeback)
1674                 seq_printf(m, " writeback=%lu", md->writeback);
1675
1676         for_each_node_state(nid, N_MEMORY)
1677                 if (md->node[nid])
1678                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1679
1680         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1681 out:
1682         seq_putc(m, '\n');
1683         m_cache_vma(m, vma);
1684         return 0;
1685 }
1686
1687 static int show_pid_numa_map(struct seq_file *m, void *v)
1688 {
1689         return show_numa_map(m, v, 1);
1690 }
1691
1692 static int show_tid_numa_map(struct seq_file *m, void *v)
1693 {
1694         return show_numa_map(m, v, 0);
1695 }
1696
1697 static const struct seq_operations proc_pid_numa_maps_op = {
1698         .start  = m_start,
1699         .next   = m_next,
1700         .stop   = m_stop,
1701         .show   = show_pid_numa_map,
1702 };
1703
1704 static const struct seq_operations proc_tid_numa_maps_op = {
1705         .start  = m_start,
1706         .next   = m_next,
1707         .stop   = m_stop,
1708         .show   = show_tid_numa_map,
1709 };
1710
1711 static int numa_maps_open(struct inode *inode, struct file *file,
1712                           const struct seq_operations *ops)
1713 {
1714         return proc_maps_open(inode, file, ops,
1715                                 sizeof(struct numa_maps_private));
1716 }
1717
1718 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1719 {
1720         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1721 }
1722
1723 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1724 {
1725         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1726 }
1727
1728 const struct file_operations proc_pid_numa_maps_operations = {
1729         .open           = pid_numa_maps_open,
1730         .read           = seq_read,
1731         .llseek         = seq_lseek,
1732         .release        = proc_map_release,
1733 };
1734
1735 const struct file_operations proc_tid_numa_maps_operations = {
1736         .open           = tid_numa_maps_open,
1737         .read           = seq_read,
1738         .llseek         = seq_lseek,
1739         .release        = proc_map_release,
1740 };
1741 #endif /* CONFIG_NUMA */