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