fs/proc/task_mmu.c

   1 #include <linux/mm.h>
   2 #include <linux/vmacache.h>
   3 #include <linux/hugetlb.h>
   4 #include <linux/huge_mm.h>
   5 #include <linux/mount.h>
   6 #include <linux/seq_file.h>
   7 #include <linux/highmem.h>
   8 #include <linux/ptrace.h>
   9 #include <linux/slab.h>
  10 #include <linux/pagemap.h>
  11 #include <linux/mempolicy.h>
  12 #include <linux/rmap.h>
  13 #include <linux/swap.h>
  14 #include <linux/swapops.h>
  15 #include <linux/mmu_notifier.h>
  16 #include <linux/page_idle.h>
  17
  18 #include <asm/elf.h>
  19 #include <asm/uaccess.h>
  20 #include <asm/tlbflush.h>
  21 #include "internal.h"
  22
  23 void task_mem(struct seq_file *m, struct mm_struct *mm)
  24 {
  25         unsigned long data, text, lib, swap, ptes, pmds;
  26         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  27
  28         /*
  29          * Note: to minimize their overhead, mm maintains hiwater_vm and
  30          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  31          * collector of these hiwater stats must therefore get total_vm
  32          * and rss too, which will usually be the higher.  Barriers? not
  33          * worth the effort, such snapshots can always be inconsistent.
  34          */
  35         hiwater_vm = total_vm = mm->total_vm;
  36         if (hiwater_vm < mm->hiwater_vm)
  37                 hiwater_vm = mm->hiwater_vm;
  38         hiwater_rss = total_rss = get_mm_rss(mm);
  39         if (hiwater_rss < mm->hiwater_rss)
  40                 hiwater_rss = mm->hiwater_rss;
  41
  42         data = mm->total_vm - mm->shared_vm - mm->stack_vm;
  43         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  44         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  45         swap = get_mm_counter(mm, MM_SWAPENTS);
  46         ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
  47         pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
  48         seq_printf(m,
  49                 "VmPeak:\t%8lu kB\n"
  50                 "VmSize:\t%8lu kB\n"
  51                 "VmLck:\t%8lu kB\n"
  52                 "VmPin:\t%8lu kB\n"
  53                 "VmHWM:\t%8lu kB\n"
  54                 "VmRSS:\t%8lu kB\n"
  55                 "VmData:\t%8lu kB\n"
  56                 "VmStk:\t%8lu kB\n"
  57                 "VmExe:\t%8lu kB\n"
  58                 "VmLib:\t%8lu kB\n"
  59                 "VmPTE:\t%8lu kB\n"
  60                 "VmPMD:\t%8lu kB\n"
  61                 "VmSwap:\t%8lu kB\n",
  62                 hiwater_vm << (PAGE_SHIFT-10),
  63                 total_vm << (PAGE_SHIFT-10),
  64                 mm->locked_vm << (PAGE_SHIFT-10),
  65                 mm->pinned_vm << (PAGE_SHIFT-10),
  66                 hiwater_rss << (PAGE_SHIFT-10),
  67                 total_rss << (PAGE_SHIFT-10),
  68                 data << (PAGE_SHIFT-10),
  69                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  70                 ptes >> 10,
  71                 pmds >> 10,
  72                 swap << (PAGE_SHIFT-10));
  73         hugetlb_report_usage(m, mm);
  74 }
  75
  76 unsigned long task_vsize(struct mm_struct *mm)
  77 {
  78         return PAGE_SIZE * mm->total_vm;
  79 }
  80
  81 unsigned long task_statm(struct mm_struct *mm,
  82                          unsigned long *shared, unsigned long *text,
  83                          unsigned long *data, unsigned long *resident)
  84 {
  85         *shared = get_mm_counter(mm, MM_FILEPAGES);
  86         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  87                                                                 >> PAGE_SHIFT;
  88         *data = mm->total_vm - mm->shared_vm;
  89         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
  90         return mm->total_vm;
  91 }
  92
  93 #ifdef CONFIG_NUMA
  94 /*
  95  * Save get_task_policy() for show_numa_map().
  96  */
  97 static void hold_task_mempolicy(struct proc_maps_private *priv)
  98 {
  99         struct task_struct *task = priv->task;
 100
 101         task_lock(task);
 102         priv->task_mempolicy = get_task_policy(task);
 103         mpol_get(priv->task_mempolicy);
 104         task_unlock(task);
 105 }
 106 static void release_task_mempolicy(struct proc_maps_private *priv)
 107 {
 108         mpol_put(priv->task_mempolicy);
 109 }
 110 #else
 111 static void hold_task_mempolicy(struct proc_maps_private *priv)
 112 {
 113 }
 114 static void release_task_mempolicy(struct proc_maps_private *priv)
 115 {
 116 }
 117 #endif
 118
 119 static void vma_stop(struct proc_maps_private *priv)
 120 {
 121         struct mm_struct *mm = priv->mm;
 122
 123         release_task_mempolicy(priv);
 124         up_read(&mm->mmap_sem);
 125         mmput(mm);
 126 }
 127
 128 static struct vm_area_struct *
 129 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
 130 {
 131         if (vma == priv->tail_vma)
 132                 return NULL;
 133         return vma->vm_next ?: priv->tail_vma;
 134 }
 135
 136 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
 137 {
 138         if (m->count < m->size) /* vma is copied successfully */
 139                 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
 140 }
 141
 142 static void *m_start(struct seq_file *m, loff_t *ppos)
 143 {
 144         struct proc_maps_private *priv = m->private;
 145         unsigned long last_addr = m->version;
 146         struct mm_struct *mm;
 147         struct vm_area_struct *vma;
 148         unsigned int pos = *ppos;
 149
 150         /* See m_cache_vma(). Zero at the start or after lseek. */
 151         if (last_addr == -1UL)
 152                 return NULL;
 153
 154         priv->task = get_proc_task(priv->inode);
 155         if (!priv->task)
 156                 return ERR_PTR(-ESRCH);
 157
 158         mm = priv->mm;
 159         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
 160                 return NULL;
 161
 162         down_read(&mm->mmap_sem);
 163         hold_task_mempolicy(priv);
 164         priv->tail_vma = get_gate_vma(mm);
 165
 166         if (last_addr) {
 167                 vma = find_vma(mm, last_addr);
 168                 if (vma && (vma = m_next_vma(priv, vma)))
 169                         return vma;
 170         }
 171
 172         m->version = 0;
 173         if (pos < mm->map_count) {
 174                 for (vma = mm->mmap; pos; pos--) {
 175                         m->version = vma->vm_start;
 176                         vma = vma->vm_next;
 177                 }
 178                 return vma;
 179         }
 180
 181         /* we do not bother to update m->version in this case */
 182         if (pos == mm->map_count && priv->tail_vma)
 183                 return priv->tail_vma;
 184
 185         vma_stop(priv);
 186         return NULL;
 187 }
 188
 189 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 190 {
 191         struct proc_maps_private *priv = m->private;
 192         struct vm_area_struct *next;
 193
 194         (*pos)++;
 195         next = m_next_vma(priv, v);
 196         if (!next)
 197                 vma_stop(priv);
 198         return next;
 199 }
 200
 201 static void m_stop(struct seq_file *m, void *v)
 202 {
 203         struct proc_maps_private *priv = m->private;
 204
 205         if (!IS_ERR_OR_NULL(v))
 206                 vma_stop(priv);
 207         if (priv->task) {
 208                 put_task_struct(priv->task);
 209                 priv->task = NULL;
 210         }
 211 }
 212
 213 static int proc_maps_open(struct inode *inode, struct file *file,
 214                         const struct seq_operations *ops, int psize)
 215 {
 216         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 217
 218         if (!priv)
 219                 return -ENOMEM;
 220
 221         priv->inode = inode;
 222         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 223         if (IS_ERR(priv->mm)) {
 224                 int err = PTR_ERR(priv->mm);
 225
 226                 seq_release_private(inode, file);
 227                 return err;
 228         }
 229
 230         return 0;
 231 }
 232
 233 static int proc_map_release(struct inode *inode, struct file *file)
 234 {
 235         struct seq_file *seq = file->private_data;
 236         struct proc_maps_private *priv = seq->private;
 237
 238         if (priv->mm)
 239                 mmdrop(priv->mm);
 240
 241         return seq_release_private(inode, file);
 242 }
 243
 244 static int do_maps_open(struct inode *inode, struct file *file,
 245                         const struct seq_operations *ops)
 246 {
 247         return proc_maps_open(inode, file, ops,
 248                                 sizeof(struct proc_maps_private));
 249 }
 250
 251 /*
 252  * Indicate if the VMA is a stack for the given task; for
 253  * /proc/PID/maps that is the stack of the main task.
 254  */
 255 static int is_stack(struct proc_maps_private *priv,
 256                     struct vm_area_struct *vma)
 257 {
 258         /*
 259          * We make no effort to guess what a given thread considers to be
 260          * its "stack".  It's not even well-defined for programs written
 261          * languages like Go.
 262          */
 263         return vma->vm_start <= vma->vm_mm->start_stack &&
 264                 vma->vm_end >= vma->vm_mm->start_stack;
 265 }
 266
 267 static void
 268 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 269 {
 270         struct mm_struct *mm = vma->vm_mm;
 271         struct file *file = vma->vm_file;
 272         struct proc_maps_private *priv = m->private;
 273         vm_flags_t flags = vma->vm_flags;
 274         unsigned long ino = 0;
 275         unsigned long long pgoff = 0;
 276         unsigned long start, end;
 277         dev_t dev = 0;
 278         const char *name = NULL;
 279
 280         if (file) {
 281                 struct inode *inode = file_inode(vma->vm_file);
 282                 dev = inode->i_sb->s_dev;
 283                 ino = inode->i_ino;
 284                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 285         }
 286
 287         /* We don't show the stack guard page in /proc/maps */
 288         start = vma->vm_start;
 289         end = vma->vm_end;
 290
 291         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 292         seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
 293                         start,
 294                         end,
 295                         flags & VM_READ ? 'r' : '-',
 296                         flags & VM_WRITE ? 'w' : '-',
 297                         flags & VM_EXEC ? 'x' : '-',
 298                         flags & VM_MAYSHARE ? 's' : 'p',
 299                         pgoff,
 300                         MAJOR(dev), MINOR(dev), ino);
 301
 302         /*
 303          * Print the dentry name for named mappings, and a
 304          * special [heap] marker for the heap:
 305          */
 306         if (file) {
 307                 seq_pad(m, ' ');
 308                 seq_file_path(m, file, "\n");
 309                 goto done;
 310         }
 311
 312         if (vma->vm_ops && vma->vm_ops->name) {
 313                 name = vma->vm_ops->name(vma);
 314                 if (name)
 315                         goto done;
 316         }
 317
 318         name = arch_vma_name(vma);
 319         if (!name) {
 320                 if (!mm) {
 321                         name = "[vdso]";
 322                         goto done;
 323                 }
 324
 325                 if (vma->vm_start <= mm->brk &&
 326                     vma->vm_end >= mm->start_brk) {
 327                         name = "[heap]";
 328                         goto done;
 329                 }
 330
 331                 if (is_stack(priv, vma))
 332                         name = "[stack]";
 333         }
 334
 335 done:
 336         if (name) {
 337                 seq_pad(m, ' ');
 338                 seq_puts(m, name);
 339         }
 340         seq_putc(m, '\n');
 341 }
 342
 343 static int show_map(struct seq_file *m, void *v, int is_pid)
 344 {
 345         show_map_vma(m, v, is_pid);
 346         m_cache_vma(m, v);
 347         return 0;
 348 }
 349
 350 static int show_pid_map(struct seq_file *m, void *v)
 351 {
 352         return show_map(m, v, 1);
 353 }
 354
 355 static int show_tid_map(struct seq_file *m, void *v)
 356 {
 357         return show_map(m, v, 0);
 358 }
 359
 360 static const struct seq_operations proc_pid_maps_op = {
 361         .start  = m_start,
 362         .next   = m_next,
 363         .stop   = m_stop,
 364         .show   = show_pid_map
 365 };
 366
 367 static const struct seq_operations proc_tid_maps_op = {
 368         .start  = m_start,
 369         .next   = m_next,
 370         .stop   = m_stop,
 371         .show   = show_tid_map
 372 };
 373
 374 static int pid_maps_open(struct inode *inode, struct file *file)
 375 {
 376         return do_maps_open(inode, file, &proc_pid_maps_op);
 377 }
 378
 379 static int tid_maps_open(struct inode *inode, struct file *file)
 380 {
 381         return do_maps_open(inode, file, &proc_tid_maps_op);
 382 }
 383
 384 const struct file_operations proc_pid_maps_operations = {
 385         .open           = pid_maps_open,
 386         .read           = seq_read,
 387         .llseek         = seq_lseek,
 388         .release        = proc_map_release,
 389 };
 390
 391 const struct file_operations proc_tid_maps_operations = {
 392         .open           = tid_maps_open,
 393         .read           = seq_read,
 394         .llseek         = seq_lseek,
 395         .release        = proc_map_release,
 396 };
 397
 398 /*
 399  * Proportional Set Size(PSS): my share of RSS.
 400  *
 401  * PSS of a process is the count of pages it has in memory, where each
 402  * page is divided by the number of processes sharing it.  So if a
 403  * process has 1000 pages all to itself, and 1000 shared with one other
 404  * process, its PSS will be 1500.
 405  *
 406  * To keep (accumulated) division errors low, we adopt a 64bit
 407  * fixed-point pss counter to minimize division errors. So (pss >>
 408  * PSS_SHIFT) would be the real byte count.
 409  *
 410  * A shift of 12 before division means (assuming 4K page size):
 411  *      - 1M 3-user-pages add up to 8KB errors;
 412  *      - supports mapcount up to 2^24, or 16M;
 413  *      - supports PSS up to 2^52 bytes, or 4PB.
 414  */
 415 #define PSS_SHIFT 12
 416
 417 #ifdef CONFIG_PROC_PAGE_MONITOR
 418 struct mem_size_stats {
 419         unsigned long resident;
 420         unsigned long shared_clean;
 421         unsigned long shared_dirty;
 422         unsigned long private_clean;
 423         unsigned long private_dirty;
 424         unsigned long referenced;
 425         unsigned long anonymous;
 426         unsigned long anonymous_thp;
 427         unsigned long swap;
 428         unsigned long shared_hugetlb;
 429         unsigned long private_hugetlb;
 430         u64 pss;
 431         u64 swap_pss;
 432 };
 433
 434 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 435                 unsigned long size, bool young, bool dirty)
 436 {
 437         int mapcount;
 438
 439         if (PageAnon(page))
 440                 mss->anonymous += size;
 441
 442         mss->resident += size;
 443         /* Accumulate the size in pages that have been accessed. */
 444         if (young || page_is_young(page) || PageReferenced(page))
 445                 mss->referenced += size;
 446         mapcount = page_mapcount(page);
 447         if (mapcount >= 2) {
 448                 u64 pss_delta;
 449
 450                 if (dirty || PageDirty(page))
 451                         mss->shared_dirty += size;
 452                 else
 453                         mss->shared_clean += size;
 454                 pss_delta = (u64)size << PSS_SHIFT;
 455                 do_div(pss_delta, mapcount);
 456                 mss->pss += pss_delta;
 457         } else {
 458                 if (dirty || PageDirty(page))
 459                         mss->private_dirty += size;
 460                 else
 461                         mss->private_clean += size;
 462                 mss->pss += (u64)size << PSS_SHIFT;
 463         }
 464 }
 465
 466 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 467                 struct mm_walk *walk)
 468 {
 469         struct mem_size_stats *mss = walk->private;
 470         struct vm_area_struct *vma = walk->vma;
 471         struct page *page = NULL;
 472
 473         if (pte_present(*pte)) {
 474                 page = vm_normal_page(vma, addr, *pte);
 475         } else if (is_swap_pte(*pte)) {
 476                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 477
 478                 if (!non_swap_entry(swpent)) {
 479                         int mapcount;
 480
 481                         mss->swap += PAGE_SIZE;
 482                         mapcount = swp_swapcount(swpent);
 483                         if (mapcount >= 2) {
 484                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 485
 486                                 do_div(pss_delta, mapcount);
 487                                 mss->swap_pss += pss_delta;
 488                         } else {
 489                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 490                         }
 491                 } else if (is_migration_entry(swpent))
 492                         page = migration_entry_to_page(swpent);
 493         }
 494
 495         if (!page)
 496                 return;
 497         smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte));
 498 }
 499
 500 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 501 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 502                 struct mm_walk *walk)
 503 {
 504         struct mem_size_stats *mss = walk->private;
 505         struct vm_area_struct *vma = walk->vma;
 506         struct page *page;
 507
 508         /* FOLL_DUMP will return -EFAULT on huge zero page */
 509         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 510         if (IS_ERR_OR_NULL(page))
 511                 return;
 512         mss->anonymous_thp += HPAGE_PMD_SIZE;
 513         smaps_account(mss, page, HPAGE_PMD_SIZE,
 514                         pmd_young(*pmd), pmd_dirty(*pmd));
 515 }
 516 #else
 517 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 518                 struct mm_walk *walk)
 519 {
 520 }
 521 #endif
 522
 523 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 524                            struct mm_walk *walk)
 525 {
 526         struct vm_area_struct *vma = walk->vma;
 527         pte_t *pte;
 528         spinlock_t *ptl;
 529
 530         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
 531                 smaps_pmd_entry(pmd, addr, walk);
 532                 spin_unlock(ptl);
 533                 return 0;
 534         }
 535
 536         if (pmd_trans_unstable(pmd))
 537                 return 0;
 538         /*
 539          * The mmap_sem held all the way back in m_start() is what
 540          * keeps khugepaged out of here and from collapsing things
 541          * in here.
 542          */
 543         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 544         for (; addr != end; pte++, addr += PAGE_SIZE)
 545                 smaps_pte_entry(pte, addr, walk);
 546         pte_unmap_unlock(pte - 1, ptl);
 547         cond_resched();
 548         return 0;
 549 }
 550
 551 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 552 {
 553         /*
 554          * Don't forget to update Documentation/ on changes.
 555          */
 556         static const char mnemonics[BITS_PER_LONG][2] = {
 557                 /*
 558                  * In case if we meet a flag we don't know about.
 559                  */
 560                 [0 ... (BITS_PER_LONG-1)] = "??",
 561
 562                 [ilog2(VM_READ)]        = "rd",
 563                 [ilog2(VM_WRITE)]       = "wr",
 564                 [ilog2(VM_EXEC)]        = "ex",
 565                 [ilog2(VM_SHARED)]      = "sh",
 566                 [ilog2(VM_MAYREAD)]     = "mr",
 567                 [ilog2(VM_MAYWRITE)]    = "mw",
 568                 [ilog2(VM_MAYEXEC)]     = "me",
 569                 [ilog2(VM_MAYSHARE)]    = "ms",
 570                 [ilog2(VM_GROWSDOWN)]   = "gd",
 571                 [ilog2(VM_PFNMAP)]      = "pf",
 572                 [ilog2(VM_DENYWRITE)]   = "dw",
 573 #ifdef CONFIG_X86_INTEL_MPX
 574                 [ilog2(VM_MPX)]         = "mp",
 575 #endif
 576                 [ilog2(VM_LOCKED)]      = "lo",
 577                 [ilog2(VM_IO)]          = "io",
 578                 [ilog2(VM_SEQ_READ)]    = "sr",
 579                 [ilog2(VM_RAND_READ)]   = "rr",
 580                 [ilog2(VM_DONTCOPY)]    = "dc",
 581                 [ilog2(VM_DONTEXPAND)]  = "de",
 582                 [ilog2(VM_ACCOUNT)]     = "ac",
 583                 [ilog2(VM_NORESERVE)]   = "nr",
 584                 [ilog2(VM_HUGETLB)]     = "ht",
 585                 [ilog2(VM_ARCH_1)]      = "ar",
 586                 [ilog2(VM_DONTDUMP)]    = "dd",
 587 #ifdef CONFIG_MEM_SOFT_DIRTY
 588                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 589 #endif
 590                 [ilog2(VM_MIXEDMAP)]    = "mm",
 591                 [ilog2(VM_HUGEPAGE)]    = "hg",
 592                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 593                 [ilog2(VM_MERGEABLE)]   = "mg",
 594                 [ilog2(VM_UFFD_MISSING)]= "um",
 595                 [ilog2(VM_UFFD_WP)]     = "uw",
 596         };
 597         size_t i;
 598
 599         seq_puts(m, "VmFlags: ");
 600         for (i = 0; i < BITS_PER_LONG; i++) {
 601                 if (vma->vm_flags & (1UL << i)) {
 602                         seq_printf(m, "%c%c ",
 603                                    mnemonics[i][0], mnemonics[i][1]);
 604                 }
 605         }
 606         seq_putc(m, '\n');
 607 }
 608
 609 #ifdef CONFIG_HUGETLB_PAGE
 610 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 611                                  unsigned long addr, unsigned long end,
 612                                  struct mm_walk *walk)
 613 {
 614         struct mem_size_stats *mss = walk->private;
 615         struct vm_area_struct *vma = walk->vma;
 616         struct page *page = NULL;
 617
 618         if (pte_present(*pte)) {
 619                 page = vm_normal_page(vma, addr, *pte);
 620         } else if (is_swap_pte(*pte)) {
 621                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 622
 623                 if (is_migration_entry(swpent))
 624                         page = migration_entry_to_page(swpent);
 625         }
 626         if (page) {
 627                 int mapcount = page_mapcount(page);
 628
 629                 if (mapcount >= 2)
 630                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 631                 else
 632                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 633         }
 634         return 0;
 635 }
 636 #endif /* HUGETLB_PAGE */
 637
 638 static int show_smap(struct seq_file *m, void *v, int is_pid)
 639 {
 640         struct vm_area_struct *vma = v;
 641         struct mem_size_stats mss;
 642         struct mm_walk smaps_walk = {
 643                 .pmd_entry = smaps_pte_range,
 644 #ifdef CONFIG_HUGETLB_PAGE
 645                 .hugetlb_entry = smaps_hugetlb_range,
 646 #endif
 647                 .mm = vma->vm_mm,
 648                 .private = &mss,
 649         };
 650
 651         memset(&mss, 0, sizeof mss);
 652         /* mmap_sem is held in m_start */
 653         walk_page_vma(vma, &smaps_walk);
 654
 655         show_map_vma(m, vma, is_pid);
 656
 657         seq_printf(m,
 658                    "Size:           %8lu kB\n"
 659                    "Rss:            %8lu kB\n"
 660                    "Pss:            %8lu kB\n"
 661                    "Shared_Clean:   %8lu kB\n"
 662                    "Shared_Dirty:   %8lu kB\n"
 663                    "Private_Clean:  %8lu kB\n"
 664                    "Private_Dirty:  %8lu kB\n"
 665                    "Referenced:     %8lu kB\n"
 666                    "Anonymous:      %8lu kB\n"
 667                    "AnonHugePages:  %8lu kB\n"
 668                    "Shared_Hugetlb: %8lu kB\n"
 669                    "Private_Hugetlb: %7lu kB\n"
 670                    "Swap:           %8lu kB\n"
 671                    "SwapPss:        %8lu kB\n"
 672                    "KernelPageSize: %8lu kB\n"
 673                    "MMUPageSize:    %8lu kB\n"
 674                    "Locked:         %8lu kB\n",
 675                    (vma->vm_end - vma->vm_start) >> 10,
 676                    mss.resident >> 10,
 677                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 678                    mss.shared_clean  >> 10,
 679                    mss.shared_dirty  >> 10,
 680                    mss.private_clean >> 10,
 681                    mss.private_dirty >> 10,
 682                    mss.referenced >> 10,
 683                    mss.anonymous >> 10,
 684                    mss.anonymous_thp >> 10,
 685                    mss.shared_hugetlb >> 10,
 686                    mss.private_hugetlb >> 10,
 687                    mss.swap >> 10,
 688                    (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
 689                    vma_kernel_pagesize(vma) >> 10,
 690                    vma_mmu_pagesize(vma) >> 10,
 691                    (vma->vm_flags & VM_LOCKED) ?
 692                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 693
 694         show_smap_vma_flags(m, vma);
 695         m_cache_vma(m, vma);
 696         return 0;
 697 }
 698
 699 static int show_pid_smap(struct seq_file *m, void *v)
 700 {
 701         return show_smap(m, v, 1);
 702 }
 703
 704 static int show_tid_smap(struct seq_file *m, void *v)
 705 {
 706         return show_smap(m, v, 0);
 707 }
 708
 709 static const struct seq_operations proc_pid_smaps_op = {
 710         .start  = m_start,
 711         .next   = m_next,
 712         .stop   = m_stop,
 713         .show   = show_pid_smap
 714 };
 715
 716 static const struct seq_operations proc_tid_smaps_op = {
 717         .start  = m_start,
 718         .next   = m_next,
 719         .stop   = m_stop,
 720         .show   = show_tid_smap
 721 };
 722
 723 static int pid_smaps_open(struct inode *inode, struct file *file)
 724 {
 725         return do_maps_open(inode, file, &proc_pid_smaps_op);
 726 }
 727
 728 static int tid_smaps_open(struct inode *inode, struct file *file)
 729 {
 730         return do_maps_open(inode, file, &proc_tid_smaps_op);
 731 }
 732
 733 const struct file_operations proc_pid_smaps_operations = {
 734         .open           = pid_smaps_open,
 735         .read           = seq_read,
 736         .llseek         = seq_lseek,
 737         .release        = proc_map_release,
 738 };
 739
 740 const struct file_operations proc_tid_smaps_operations = {
 741         .open           = tid_smaps_open,
 742         .read           = seq_read,
 743         .llseek         = seq_lseek,
 744         .release        = proc_map_release,
 745 };
 746
 747 enum clear_refs_types {
 748         CLEAR_REFS_ALL = 1,
 749         CLEAR_REFS_ANON,
 750         CLEAR_REFS_MAPPED,
 751         CLEAR_REFS_SOFT_DIRTY,
 752         CLEAR_REFS_MM_HIWATER_RSS,
 753         CLEAR_REFS_LAST,
 754 };
 755
 756 struct clear_refs_private {
 757         enum clear_refs_types type;
 758 };
 759
 760 #ifdef CONFIG_MEM_SOFT_DIRTY
 761 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 762                 unsigned long addr, pte_t *pte)
 763 {
 764         /*
 765          * The soft-dirty tracker uses #PF-s to catch writes
 766          * to pages, so write-protect the pte as well. See the
 767          * Documentation/vm/soft-dirty.txt for full description
 768          * of how soft-dirty works.
 769          */
 770         pte_t ptent = *pte;
 771
 772         if (pte_present(ptent)) {
 773                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 774                 ptent = pte_wrprotect(ptent);
 775                 ptent = pte_clear_soft_dirty(ptent);
 776                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 777         } else if (is_swap_pte(ptent)) {
 778                 ptent = pte_swp_clear_soft_dirty(ptent);
 779                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 780         }
 781 }
 782 #else
 783 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 784                 unsigned long addr, pte_t *pte)
 785 {
 786 }
 787 #endif
 788
 789 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 790 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 791                 unsigned long addr, pmd_t *pmdp)
 792 {
 793         pmd_t pmd = *pmdp;
 794
 795         /* See comment in change_huge_pmd() */
 796         pmdp_invalidate(vma, addr, pmdp);
 797         if (pmd_dirty(*pmdp))
 798                 pmd = pmd_mkdirty(pmd);
 799         if (pmd_young(*pmdp))
 800                 pmd = pmd_mkyoung(pmd);
 801
 802         pmd = pmd_wrprotect(pmd);
 803         pmd = pmd_clear_soft_dirty(pmd);
 804
 805         if (vma->vm_flags & VM_SOFTDIRTY)
 806                 vma->vm_flags &= ~VM_SOFTDIRTY;
 807
 808         set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 809 }
 810 #else
 811 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 812                 unsigned long addr, pmd_t *pmdp)
 813 {
 814 }
 815 #endif
 816
 817 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 818                                 unsigned long end, struct mm_walk *walk)
 819 {
 820         struct clear_refs_private *cp = walk->private;
 821         struct vm_area_struct *vma = walk->vma;
 822         pte_t *pte, ptent;
 823         spinlock_t *ptl;
 824         struct page *page;
 825
 826         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
 827                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 828                         clear_soft_dirty_pmd(vma, addr, pmd);
 829                         goto out;
 830                 }
 831
 832                 page = pmd_page(*pmd);
 833
 834                 /* Clear accessed and referenced bits. */
 835                 pmdp_test_and_clear_young(vma, addr, pmd);
 836                 test_and_clear_page_young(page);
 837                 ClearPageReferenced(page);
 838 out:
 839                 spin_unlock(ptl);
 840                 return 0;
 841         }
 842
 843         if (pmd_trans_unstable(pmd))
 844                 return 0;
 845
 846         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 847         for (; addr != end; pte++, addr += PAGE_SIZE) {
 848                 ptent = *pte;
 849
 850                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 851                         clear_soft_dirty(vma, addr, pte);
 852                         continue;
 853                 }
 854
 855                 if (!pte_present(ptent))
 856                         continue;
 857
 858                 page = vm_normal_page(vma, addr, ptent);
 859                 if (!page)
 860                         continue;
 861
 862                 /* Clear accessed and referenced bits. */
 863                 ptep_test_and_clear_young(vma, addr, pte);
 864                 test_and_clear_page_young(page);
 865                 ClearPageReferenced(page);
 866         }
 867         pte_unmap_unlock(pte - 1, ptl);
 868         cond_resched();
 869         return 0;
 870 }
 871
 872 static int clear_refs_test_walk(unsigned long start, unsigned long end,
 873                                 struct mm_walk *walk)
 874 {
 875         struct clear_refs_private *cp = walk->private;
 876         struct vm_area_struct *vma = walk->vma;
 877
 878         if (vma->vm_flags & VM_PFNMAP)
 879                 return 1;
 880
 881         /*
 882          * Writing 1 to /proc/pid/clear_refs affects all pages.
 883          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
 884          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
 885          * Writing 4 to /proc/pid/clear_refs affects all pages.
 886          */
 887         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
 888                 return 1;
 889         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
 890                 return 1;
 891         return 0;
 892 }
 893
 894 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 895                                 size_t count, loff_t *ppos)
 896 {
 897         struct task_struct *task;
 898         char buffer[PROC_NUMBUF];
 899         struct mm_struct *mm;
 900         struct vm_area_struct *vma;
 901         enum clear_refs_types type;
 902         int itype;
 903         int rv;
 904
 905         memset(buffer, 0, sizeof(buffer));
 906         if (count > sizeof(buffer) - 1)
 907                 count = sizeof(buffer) - 1;
 908         if (copy_from_user(buffer, buf, count))
 909                 return -EFAULT;
 910         rv = kstrtoint(strstrip(buffer), 10, &itype);
 911         if (rv < 0)
 912                 return rv;
 913         type = (enum clear_refs_types)itype;
 914         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
 915                 return -EINVAL;
 916
 917         task = get_proc_task(file_inode(file));
 918         if (!task)
 919                 return -ESRCH;
 920         mm = get_task_mm(task);
 921         if (mm) {
 922                 struct clear_refs_private cp = {
 923                         .type = type,
 924                 };
 925                 struct mm_walk clear_refs_walk = {
 926                         .pmd_entry = clear_refs_pte_range,
 927                         .test_walk = clear_refs_test_walk,
 928                         .mm = mm,
 929                         .private = &cp,
 930                 };
 931
 932                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
 933                         /*
 934                          * Writing 5 to /proc/pid/clear_refs resets the peak
 935                          * resident set size to this mm's current rss value.
 936                          */
 937                         down_write(&mm->mmap_sem);
 938                         reset_mm_hiwater_rss(mm);
 939                         up_write(&mm->mmap_sem);
 940                         goto out_mm;
 941                 }
 942
 943                 down_read(&mm->mmap_sem);
 944                 if (type == CLEAR_REFS_SOFT_DIRTY) {
 945                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
 946                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
 947                                         continue;
 948                                 up_read(&mm->mmap_sem);
 949                                 down_write(&mm->mmap_sem);
 950                                 /*
 951                                  * Avoid to modify vma->vm_flags
 952                                  * without locked ops while the
 953                                  * coredump reads the vm_flags.
 954                                  */
 955                                 if (!mmget_still_valid(mm)) {
 956                                         /*
 957                                          * Silently return "count"
 958                                          * like if get_task_mm()
 959                                          * failed. FIXME: should this
 960                                          * function have returned
 961                                          * -ESRCH if get_task_mm()
 962                                          * failed like if
 963                                          * get_proc_task() fails?
 964                                          */
 965                                         up_write(&mm->mmap_sem);
 966                                         goto out_mm;
 967                                 }
 968                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
 969                                         vma->vm_flags &= ~VM_SOFTDIRTY;
 970                                         vma_set_page_prot(vma);
 971                                 }
 972                                 downgrade_write(&mm->mmap_sem);
 973                                 break;
 974                         }
 975                         mmu_notifier_invalidate_range_start(mm, 0, -1);
 976                 }
 977                 walk_page_range(0, ~0UL, &clear_refs_walk);
 978                 if (type == CLEAR_REFS_SOFT_DIRTY)
 979                         mmu_notifier_invalidate_range_end(mm, 0, -1);
 980                 flush_tlb_mm(mm);
 981                 up_read(&mm->mmap_sem);
 982 out_mm:
 983                 mmput(mm);
 984         }
 985         put_task_struct(task);
 986
 987         return count;
 988 }
 989
 990 const struct file_operations proc_clear_refs_operations = {
 991         .write          = clear_refs_write,
 992         .llseek         = noop_llseek,
 993 };
 994
 995 typedef struct {
 996         u64 pme;
 997 } pagemap_entry_t;
 998
 999 struct pagemapread {
1000         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1001         pagemap_entry_t *buffer;
1002         bool show_pfn;
1003 };
1004
1005 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1006 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1007
1008 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1009 #define PM_PFRAME_BITS          55
1010 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1011 #define PM_SOFT_DIRTY           BIT_ULL(55)
1012 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1013 #define PM_FILE                 BIT_ULL(61)
1014 #define PM_SWAP                 BIT_ULL(62)
1015 #define PM_PRESENT              BIT_ULL(63)
1016
1017 #define PM_END_OF_BUFFER    1
1018
1019 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1020 {
1021         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1022 }
1023
1024 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1025                           struct pagemapread *pm)
1026 {
1027         pm->buffer[pm->pos++] = *pme;
1028         if (pm->pos >= pm->len)
1029                 return PM_END_OF_BUFFER;
1030         return 0;
1031 }
1032
1033 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1034                                 struct mm_walk *walk)
1035 {
1036         struct pagemapread *pm = walk->private;
1037         unsigned long addr = start;
1038         int err = 0;
1039
1040         while (addr < end) {
1041                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1042                 pagemap_entry_t pme = make_pme(0, 0);
1043                 /* End of address space hole, which we mark as non-present. */
1044                 unsigned long hole_end;
1045
1046                 if (vma)
1047                         hole_end = min(end, vma->vm_start);
1048                 else
1049                         hole_end = end;
1050
1051                 for (; addr < hole_end; addr += PAGE_SIZE) {
1052                         err = add_to_pagemap(addr, &pme, pm);
1053                         if (err)
1054                                 goto out;
1055                 }
1056
1057                 if (!vma)
1058                         break;
1059
1060                 /* Addresses in the VMA. */
1061                 if (vma->vm_flags & VM_SOFTDIRTY)
1062                         pme = make_pme(0, PM_SOFT_DIRTY);
1063                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1064                         err = add_to_pagemap(addr, &pme, pm);
1065                         if (err)
1066                                 goto out;
1067                 }
1068         }
1069 out:
1070         return err;
1071 }
1072
1073 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1074                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1075 {
1076         u64 frame = 0, flags = 0;
1077         struct page *page = NULL;
1078
1079         if (pte_present(pte)) {
1080                 if (pm->show_pfn)
1081                         frame = pte_pfn(pte);
1082                 flags |= PM_PRESENT;
1083                 page = vm_normal_page(vma, addr, pte);
1084                 if (pte_soft_dirty(pte))
1085                         flags |= PM_SOFT_DIRTY;
1086         } else if (is_swap_pte(pte)) {
1087                 swp_entry_t entry;
1088                 if (pte_swp_soft_dirty(pte))
1089                         flags |= PM_SOFT_DIRTY;
1090                 entry = pte_to_swp_entry(pte);
1091                 frame = swp_type(entry) |
1092                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1093                 flags |= PM_SWAP;
1094                 if (is_migration_entry(entry))
1095                         page = migration_entry_to_page(entry);
1096         }
1097
1098         if (page && !PageAnon(page))
1099                 flags |= PM_FILE;
1100         if (page && page_mapcount(page) == 1)
1101                 flags |= PM_MMAP_EXCLUSIVE;
1102         if (vma->vm_flags & VM_SOFTDIRTY)
1103                 flags |= PM_SOFT_DIRTY;
1104
1105         return make_pme(frame, flags);
1106 }
1107
1108 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1109                              struct mm_walk *walk)
1110 {
1111         struct vm_area_struct *vma = walk->vma;
1112         struct pagemapread *pm = walk->private;
1113         spinlock_t *ptl;
1114         pte_t *pte, *orig_pte;
1115         int err = 0;
1116
1117 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1118         if (pmd_trans_huge_lock(pmdp, vma, &ptl) == 1) {
1119                 u64 flags = 0, frame = 0;
1120                 pmd_t pmd = *pmdp;
1121
1122                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1123                         flags |= PM_SOFT_DIRTY;
1124
1125                 /*
1126                  * Currently pmd for thp is always present because thp
1127                  * can not be swapped-out, migrated, or HWPOISONed
1128                  * (split in such cases instead.)
1129                  * This if-check is just to prepare for future implementation.
1130                  */
1131                 if (pmd_present(pmd)) {
1132                         struct page *page = pmd_page(pmd);
1133
1134                         if (page_mapcount(page) == 1)
1135                                 flags |= PM_MMAP_EXCLUSIVE;
1136
1137                         flags |= PM_PRESENT;
1138                         if (pm->show_pfn)
1139                                 frame = pmd_pfn(pmd) +
1140                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1141                 }
1142
1143                 for (; addr != end; addr += PAGE_SIZE) {
1144                         pagemap_entry_t pme = make_pme(frame, flags);
1145
1146                         err = add_to_pagemap(addr, &pme, pm);
1147                         if (err)
1148                                 break;
1149                         if (pm->show_pfn && (flags & PM_PRESENT))
1150                                 frame++;
1151                 }
1152                 spin_unlock(ptl);
1153                 return err;
1154         }
1155
1156         if (pmd_trans_unstable(pmdp))
1157                 return 0;
1158 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1159
1160         /*
1161          * We can assume that @vma always points to a valid one and @end never
1162          * goes beyond vma->vm_end.
1163          */
1164         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1165         for (; addr < end; pte++, addr += PAGE_SIZE) {
1166                 pagemap_entry_t pme;
1167
1168                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1169                 err = add_to_pagemap(addr, &pme, pm);
1170                 if (err)
1171                         break;
1172         }
1173         pte_unmap_unlock(orig_pte, ptl);
1174
1175         cond_resched();
1176
1177         return err;
1178 }
1179
1180 #ifdef CONFIG_HUGETLB_PAGE
1181 /* This function walks within one hugetlb entry in the single call */
1182 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1183                                  unsigned long addr, unsigned long end,
1184                                  struct mm_walk *walk)
1185 {
1186         struct pagemapread *pm = walk->private;
1187         struct vm_area_struct *vma = walk->vma;
1188         u64 flags = 0, frame = 0;
1189         int err = 0;
1190         pte_t pte;
1191
1192         if (vma->vm_flags & VM_SOFTDIRTY)
1193                 flags |= PM_SOFT_DIRTY;
1194
1195         pte = huge_ptep_get(ptep);
1196         if (pte_present(pte)) {
1197                 struct page *page = pte_page(pte);
1198
1199                 if (!PageAnon(page))
1200                         flags |= PM_FILE;
1201
1202                 if (page_mapcount(page) == 1)
1203                         flags |= PM_MMAP_EXCLUSIVE;
1204
1205                 flags |= PM_PRESENT;
1206                 if (pm->show_pfn)
1207                         frame = pte_pfn(pte) +
1208                                 ((addr & ~hmask) >> PAGE_SHIFT);
1209         }
1210
1211         for (; addr != end; addr += PAGE_SIZE) {
1212                 pagemap_entry_t pme = make_pme(frame, flags);
1213
1214                 err = add_to_pagemap(addr, &pme, pm);
1215                 if (err)
1216                         return err;
1217                 if (pm->show_pfn && (flags & PM_PRESENT))
1218                         frame++;
1219         }
1220
1221         cond_resched();
1222
1223         return err;
1224 }
1225 #endif /* HUGETLB_PAGE */
1226
1227 /*
1228  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1229  *
1230  * For each page in the address space, this file contains one 64-bit entry
1231  * consisting of the following:
1232  *
1233  * Bits 0-54  page frame number (PFN) if present
1234  * Bits 0-4   swap type if swapped
1235  * Bits 5-54  swap offset if swapped
1236  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1237  * Bit  56    page exclusively mapped
1238  * Bits 57-60 zero
1239  * Bit  61    page is file-page or shared-anon
1240  * Bit  62    page swapped
1241  * Bit  63    page present
1242  *
1243  * If the page is not present but in swap, then the PFN contains an
1244  * encoding of the swap file number and the page's offset into the
1245  * swap. Unmapped pages return a null PFN. This allows determining
1246  * precisely which pages are mapped (or in swap) and comparing mapped
1247  * pages between processes.
1248  *
1249  * Efficient users of this interface will use /proc/pid/maps to
1250  * determine which areas of memory are actually mapped and llseek to
1251  * skip over unmapped regions.
1252  */
1253 static ssize_t pagemap_read(struct file *file, char __user *buf,
1254                             size_t count, loff_t *ppos)
1255 {
1256         struct mm_struct *mm = file->private_data;
1257         struct pagemapread pm;
1258         struct mm_walk pagemap_walk = {};
1259         unsigned long src;
1260         unsigned long svpfn;
1261         unsigned long start_vaddr;
1262         unsigned long end_vaddr;
1263         int ret = 0, copied = 0;
1264
1265         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1266                 goto out;
1267
1268         ret = -EINVAL;
1269         /* file position must be aligned */
1270         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1271                 goto out_mm;
1272
1273         ret = 0;
1274         if (!count)
1275                 goto out_mm;
1276
1277         /* do not disclose physical addresses: attack vector */
1278         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1279
1280         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1281         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1282         ret = -ENOMEM;
1283         if (!pm.buffer)
1284                 goto out_mm;
1285
1286         pagemap_walk.pmd_entry = pagemap_pmd_range;
1287         pagemap_walk.pte_hole = pagemap_pte_hole;
1288 #ifdef CONFIG_HUGETLB_PAGE
1289         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1290 #endif
1291         pagemap_walk.mm = mm;
1292         pagemap_walk.private = &pm;
1293
1294         src = *ppos;
1295         svpfn = src / PM_ENTRY_BYTES;
1296         start_vaddr = svpfn << PAGE_SHIFT;
1297         end_vaddr = mm->task_size;
1298
1299         /* watch out for wraparound */
1300         if (svpfn > mm->task_size >> PAGE_SHIFT)
1301                 start_vaddr = end_vaddr;
1302
1303         /*
1304          * The odds are that this will stop walking way
1305          * before end_vaddr, because the length of the
1306          * user buffer is tracked in "pm", and the walk
1307          * will stop when we hit the end of the buffer.
1308          */
1309         ret = 0;
1310         while (count && (start_vaddr < end_vaddr)) {
1311                 int len;
1312                 unsigned long end;
1313
1314                 pm.pos = 0;
1315                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1316                 /* overflow ? */
1317                 if (end < start_vaddr || end > end_vaddr)
1318                         end = end_vaddr;
1319                 down_read(&mm->mmap_sem);
1320                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1321                 up_read(&mm->mmap_sem);
1322                 start_vaddr = end;
1323
1324                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1325                 if (copy_to_user(buf, pm.buffer, len)) {
1326                         ret = -EFAULT;
1327                         goto out_free;
1328                 }
1329                 copied += len;
1330                 buf += len;
1331                 count -= len;
1332         }
1333         *ppos += copied;
1334         if (!ret || ret == PM_END_OF_BUFFER)
1335                 ret = copied;
1336
1337 out_free:
1338         kfree(pm.buffer);
1339 out_mm:
1340         mmput(mm);
1341 out:
1342         return ret;
1343 }
1344
1345 static int pagemap_open(struct inode *inode, struct file *file)
1346 {
1347         struct mm_struct *mm;
1348
1349         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1350         if (IS_ERR(mm))
1351                 return PTR_ERR(mm);
1352         file->private_data = mm;
1353         return 0;
1354 }
1355
1356 static int pagemap_release(struct inode *inode, struct file *file)
1357 {
1358         struct mm_struct *mm = file->private_data;
1359
1360         if (mm)
1361                 mmdrop(mm);
1362         return 0;
1363 }
1364
1365 const struct file_operations proc_pagemap_operations = {
1366         .llseek         = mem_lseek, /* borrow this */
1367         .read           = pagemap_read,
1368         .open           = pagemap_open,
1369         .release        = pagemap_release,
1370 };
1371 #endif /* CONFIG_PROC_PAGE_MONITOR */
1372
1373 #ifdef CONFIG_NUMA
1374
1375 struct numa_maps {
1376         unsigned long pages;
1377         unsigned long anon;
1378         unsigned long active;
1379         unsigned long writeback;
1380         unsigned long mapcount_max;
1381         unsigned long dirty;
1382         unsigned long swapcache;
1383         unsigned long node[MAX_NUMNODES];
1384 };
1385
1386 struct numa_maps_private {
1387         struct proc_maps_private proc_maps;
1388         struct numa_maps md;
1389 };
1390
1391 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1392                         unsigned long nr_pages)
1393 {
1394         int count = page_mapcount(page);
1395
1396         md->pages += nr_pages;
1397         if (pte_dirty || PageDirty(page))
1398                 md->dirty += nr_pages;
1399
1400         if (PageSwapCache(page))
1401                 md->swapcache += nr_pages;
1402
1403         if (PageActive(page) || PageUnevictable(page))
1404                 md->active += nr_pages;
1405
1406         if (PageWriteback(page))
1407                 md->writeback += nr_pages;
1408
1409         if (PageAnon(page))
1410                 md->anon += nr_pages;
1411
1412         if (count > md->mapcount_max)
1413                 md->mapcount_max = count;
1414
1415         md->node[page_to_nid(page)] += nr_pages;
1416 }
1417
1418 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1419                 unsigned long addr)
1420 {
1421         struct page *page;
1422         int nid;
1423
1424         if (!pte_present(pte))
1425                 return NULL;
1426
1427         page = vm_normal_page(vma, addr, pte);
1428         if (!page)
1429                 return NULL;
1430
1431         if (PageReserved(page))
1432                 return NULL;
1433
1434         nid = page_to_nid(page);
1435         if (!node_isset(nid, node_states[N_MEMORY]))
1436                 return NULL;
1437
1438         return page;
1439 }
1440
1441 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1442 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1443                                               struct vm_area_struct *vma,
1444                                               unsigned long addr)
1445 {
1446         struct page *page;
1447         int nid;
1448
1449         if (!pmd_present(pmd))
1450                 return NULL;
1451
1452         page = vm_normal_page_pmd(vma, addr, pmd);
1453         if (!page)
1454                 return NULL;
1455
1456         if (PageReserved(page))
1457                 return NULL;
1458
1459         nid = page_to_nid(page);
1460         if (!node_isset(nid, node_states[N_MEMORY]))
1461                 return NULL;
1462
1463         return page;
1464 }
1465 #endif
1466
1467 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1468                 unsigned long end, struct mm_walk *walk)
1469 {
1470         struct numa_maps *md = walk->private;
1471         struct vm_area_struct *vma = walk->vma;
1472         spinlock_t *ptl;
1473         pte_t *orig_pte;
1474         pte_t *pte;
1475
1476 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1477         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1478                 struct page *page;
1479
1480                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1481                 if (page)
1482                         gather_stats(page, md, pmd_dirty(*pmd),
1483                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1484                 spin_unlock(ptl);
1485                 return 0;
1486         }
1487
1488         if (pmd_trans_unstable(pmd))
1489                 return 0;
1490 #endif
1491         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1492         do {
1493                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1494                 if (!page)
1495                         continue;
1496                 gather_stats(page, md, pte_dirty(*pte), 1);
1497
1498         } while (pte++, addr += PAGE_SIZE, addr != end);
1499         pte_unmap_unlock(orig_pte, ptl);
1500         return 0;
1501 }
1502 #ifdef CONFIG_HUGETLB_PAGE
1503 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1504                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1505 {
1506         pte_t huge_pte = huge_ptep_get(pte);
1507         struct numa_maps *md;
1508         struct page *page;
1509
1510         if (!pte_present(huge_pte))
1511                 return 0;
1512
1513         page = pte_page(huge_pte);
1514         if (!page)
1515                 return 0;
1516
1517         md = walk->private;
1518         gather_stats(page, md, pte_dirty(huge_pte), 1);
1519         return 0;
1520 }
1521
1522 #else
1523 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1524                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1525 {
1526         return 0;
1527 }
1528 #endif
1529
1530 /*
1531  * Display pages allocated per node and memory policy via /proc.
1532  */
1533 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1534 {
1535         struct numa_maps_private *numa_priv = m->private;
1536         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1537         struct vm_area_struct *vma = v;
1538         struct numa_maps *md = &numa_priv->md;
1539         struct file *file = vma->vm_file;
1540         struct mm_struct *mm = vma->vm_mm;
1541         struct mm_walk walk = {
1542                 .hugetlb_entry = gather_hugetlb_stats,
1543                 .pmd_entry = gather_pte_stats,
1544                 .private = md,
1545                 .mm = mm,
1546         };
1547         struct mempolicy *pol;
1548         char buffer[64];
1549         int nid;
1550
1551         if (!mm)
1552                 return 0;
1553
1554         /* Ensure we start with an empty set of numa_maps statistics. */
1555         memset(md, 0, sizeof(*md));
1556
1557         pol = __get_vma_policy(vma, vma->vm_start);
1558         if (pol) {
1559                 mpol_to_str(buffer, sizeof(buffer), pol);
1560                 mpol_cond_put(pol);
1561         } else {
1562                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1563         }
1564
1565         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1566
1567         if (file) {
1568                 seq_puts(m, " file=");
1569                 seq_file_path(m, file, "\n\t= ");
1570         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1571                 seq_puts(m, " heap");
1572         } else if (is_stack(proc_priv, vma)) {
1573                 seq_puts(m, " stack");
1574         }
1575
1576         if (is_vm_hugetlb_page(vma))
1577                 seq_puts(m, " huge");
1578
1579         /* mmap_sem is held by m_start */
1580         walk_page_vma(vma, &walk);
1581
1582         if (!md->pages)
1583                 goto out;
1584
1585         if (md->anon)
1586                 seq_printf(m, " anon=%lu", md->anon);
1587
1588         if (md->dirty)
1589                 seq_printf(m, " dirty=%lu", md->dirty);
1590
1591         if (md->pages != md->anon && md->pages != md->dirty)
1592                 seq_printf(m, " mapped=%lu", md->pages);
1593
1594         if (md->mapcount_max > 1)
1595                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1596
1597         if (md->swapcache)
1598                 seq_printf(m, " swapcache=%lu", md->swapcache);
1599
1600         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1601                 seq_printf(m, " active=%lu", md->active);
1602
1603         if (md->writeback)
1604                 seq_printf(m, " writeback=%lu", md->writeback);
1605
1606         for_each_node_state(nid, N_MEMORY)
1607                 if (md->node[nid])
1608                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1609
1610         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1611 out:
1612         seq_putc(m, '\n');
1613         m_cache_vma(m, vma);
1614         return 0;
1615 }
1616
1617 static int show_pid_numa_map(struct seq_file *m, void *v)
1618 {
1619         return show_numa_map(m, v, 1);
1620 }
1621
1622 static int show_tid_numa_map(struct seq_file *m, void *v)
1623 {
1624         return show_numa_map(m, v, 0);
1625 }
1626
1627 static const struct seq_operations proc_pid_numa_maps_op = {
1628         .start  = m_start,
1629         .next   = m_next,
1630         .stop   = m_stop,
1631         .show   = show_pid_numa_map,
1632 };
1633
1634 static const struct seq_operations proc_tid_numa_maps_op = {
1635         .start  = m_start,
1636         .next   = m_next,
1637         .stop   = m_stop,
1638         .show   = show_tid_numa_map,
1639 };
1640
1641 static int numa_maps_open(struct inode *inode, struct file *file,
1642                           const struct seq_operations *ops)
1643 {
1644         return proc_maps_open(inode, file, ops,
1645                                 sizeof(struct numa_maps_private));
1646 }
1647
1648 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1649 {
1650         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1651 }
1652
1653 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1654 {
1655         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1656 }
1657
1658 const struct file_operations proc_pid_numa_maps_operations = {
1659         .open           = pid_numa_maps_open,
1660         .read           = seq_read,
1661         .llseek         = seq_lseek,
1662         .release        = proc_map_release,
1663 };
1664
1665 const struct file_operations proc_tid_numa_maps_operations = {
1666         .open           = tid_numa_maps_open,
1667         .read           = seq_read,
1668         .llseek         = seq_lseek,
1669         .release        = proc_map_release,
1670 };
1671 #endif /* CONFIG_NUMA */