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