linux/fs/proc/task_mmu.c
<<
>>
Prefs
   1#include <linux/mm.h>
   2#include <linux/hugetlb.h>
   3#include <linux/mount.h>
   4#include <linux/seq_file.h>
   5#include <linux/highmem.h>
   6#include <linux/ptrace.h>
   7#include <linux/slab.h>
   8#include <linux/pagemap.h>
   9#include <linux/mempolicy.h>
  10#include <linux/swap.h>
  11#include <linux/swapops.h>
  12
  13#include <asm/elf.h>
  14#include <asm/uaccess.h>
  15#include <asm/tlbflush.h>
  16#include "internal.h"
  17
  18void task_mem(struct seq_file *m, struct mm_struct *mm)
  19{
  20        unsigned long data, text, lib, swap;
  21        unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  22
  23        /*
  24         * Note: to minimize their overhead, mm maintains hiwater_vm and
  25         * hiwater_rss only when about to *lower* total_vm or rss.  Any
  26         * collector of these hiwater stats must therefore get total_vm
  27         * and rss too, which will usually be the higher.  Barriers? not
  28         * worth the effort, such snapshots can always be inconsistent.
  29         */
  30        hiwater_vm = total_vm = mm->total_vm;
  31        if (hiwater_vm < mm->hiwater_vm)
  32                hiwater_vm = mm->hiwater_vm;
  33        hiwater_rss = total_rss = get_mm_rss(mm);
  34        if (hiwater_rss < mm->hiwater_rss)
  35                hiwater_rss = mm->hiwater_rss;
  36
  37        data = mm->total_vm - mm->shared_vm - mm->stack_vm;
  38        text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  39        lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  40        swap = get_mm_counter(mm, MM_SWAPENTS);
  41        seq_printf(m,
  42                "VmPeak:\t%8lu kB\n"
  43                "VmSize:\t%8lu kB\n"
  44                "VmLck:\t%8lu kB\n"
  45                "VmHWM:\t%8lu kB\n"
  46                "VmRSS:\t%8lu kB\n"
  47                "VmData:\t%8lu kB\n"
  48                "VmStk:\t%8lu kB\n"
  49                "VmExe:\t%8lu kB\n"
  50                "VmLib:\t%8lu kB\n"
  51                "VmPTE:\t%8lu kB\n"
  52                "VmSwap:\t%8lu kB\n",
  53                hiwater_vm << (PAGE_SHIFT-10),
  54                (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
  55                mm->locked_vm << (PAGE_SHIFT-10),
  56                hiwater_rss << (PAGE_SHIFT-10),
  57                total_rss << (PAGE_SHIFT-10),
  58                data << (PAGE_SHIFT-10),
  59                mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  60                (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
  61                swap << (PAGE_SHIFT-10));
  62}
  63
  64unsigned long task_vsize(struct mm_struct *mm)
  65{
  66        return PAGE_SIZE * mm->total_vm;
  67}
  68
  69unsigned long task_statm(struct mm_struct *mm,
  70                         unsigned long *shared, unsigned long *text,
  71                         unsigned long *data, unsigned long *resident)
  72{
  73        *shared = get_mm_counter(mm, MM_FILEPAGES);
  74        *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  75                                                                >> PAGE_SHIFT;
  76        *data = mm->total_vm - mm->shared_vm;
  77        *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
  78        return mm->total_vm;
  79}
  80
  81static void pad_len_spaces(struct seq_file *m, int len)
  82{
  83        len = 25 + sizeof(void*) * 6 - len;
  84        if (len < 1)
  85                len = 1;
  86        seq_printf(m, "%*c", len, ' ');
  87}
  88
  89static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
  90{
  91        if (vma && vma != priv->tail_vma) {
  92                struct mm_struct *mm = vma->vm_mm;
  93                up_read(&mm->mmap_sem);
  94                mmput(mm);
  95        }
  96}
  97
  98static void *m_start(struct seq_file *m, loff_t *pos)
  99{
 100        struct proc_maps_private *priv = m->private;
 101        unsigned long last_addr = m->version;
 102        struct mm_struct *mm;
 103        struct vm_area_struct *vma, *tail_vma = NULL;
 104        loff_t l = *pos;
 105
 106        /* Clear the per syscall fields in priv */
 107        priv->task = NULL;
 108        priv->tail_vma = NULL;
 109
 110        /*
 111         * We remember last_addr rather than next_addr to hit with
 112         * mmap_cache most of the time. We have zero last_addr at
 113         * the beginning and also after lseek. We will have -1 last_addr
 114         * after the end of the vmas.
 115         */
 116
 117        if (last_addr == -1UL)
 118                return NULL;
 119
 120        priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
 121        if (!priv->task)
 122                return NULL;
 123
 124        mm = mm_for_maps(priv->task);
 125        if (!mm)
 126                return NULL;
 127        down_read(&mm->mmap_sem);
 128
 129        tail_vma = get_gate_vma(priv->task);
 130        priv->tail_vma = tail_vma;
 131
 132        /* Start with last addr hint */
 133        vma = find_vma(mm, last_addr);
 134        if (last_addr && vma) {
 135                vma = vma->vm_next;
 136                goto out;
 137        }
 138
 139        /*
 140         * Check the vma index is within the range and do
 141         * sequential scan until m_index.
 142         */
 143        vma = NULL;
 144        if ((unsigned long)l < mm->map_count) {
 145                vma = mm->mmap;
 146                while (l-- && vma)
 147                        vma = vma->vm_next;
 148                goto out;
 149        }
 150
 151        if (l != mm->map_count)
 152                tail_vma = NULL; /* After gate vma */
 153
 154out:
 155        if (vma)
 156                return vma;
 157
 158        /* End of vmas has been reached */
 159        m->version = (tail_vma != NULL)? 0: -1UL;
 160        up_read(&mm->mmap_sem);
 161        mmput(mm);
 162        return tail_vma;
 163}
 164
 165static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 166{
 167        struct proc_maps_private *priv = m->private;
 168        struct vm_area_struct *vma = v;
 169        struct vm_area_struct *tail_vma = priv->tail_vma;
 170
 171        (*pos)++;
 172        if (vma && (vma != tail_vma) && vma->vm_next)
 173                return vma->vm_next;
 174        vma_stop(priv, vma);
 175        return (vma != tail_vma)? tail_vma: NULL;
 176}
 177
 178static void m_stop(struct seq_file *m, void *v)
 179{
 180        struct proc_maps_private *priv = m->private;
 181        struct vm_area_struct *vma = v;
 182
 183        vma_stop(priv, vma);
 184        if (priv->task)
 185                put_task_struct(priv->task);
 186}
 187
 188static int do_maps_open(struct inode *inode, struct file *file,
 189                        const struct seq_operations *ops)
 190{
 191        struct proc_maps_private *priv;
 192        int ret = -ENOMEM;
 193        priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 194        if (priv) {
 195                priv->pid = proc_pid(inode);
 196                ret = seq_open(file, ops);
 197                if (!ret) {
 198                        struct seq_file *m = file->private_data;
 199                        m->private = priv;
 200                } else {
 201                        kfree(priv);
 202                }
 203        }
 204        return ret;
 205}
 206
 207static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 208{
 209        struct mm_struct *mm = vma->vm_mm;
 210        struct file *file = vma->vm_file;
 211        int flags = vma->vm_flags;
 212        unsigned long ino = 0;
 213        unsigned long long pgoff = 0;
 214        unsigned long start, end;
 215        dev_t dev = 0;
 216        int len;
 217
 218        if (file) {
 219                struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
 220                dev = inode->i_sb->s_dev;
 221                ino = inode->i_ino;
 222                pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 223        }
 224
 225        /* We don't show the stack guard page in /proc/maps */
 226        start = vma->vm_start;
 227        if (stack_guard_page_start(vma, start))
 228                start += PAGE_SIZE;
 229        end = vma->vm_end;
 230        if (stack_guard_page_end(vma, end))
 231                end -= PAGE_SIZE;
 232
 233        seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
 234                        start,
 235                        end,
 236                        flags & VM_READ ? 'r' : '-',
 237                        flags & VM_WRITE ? 'w' : '-',
 238                        flags & VM_EXEC ? 'x' : '-',
 239                        flags & VM_MAYSHARE ? 's' : 'p',
 240                        pgoff,
 241                        MAJOR(dev), MINOR(dev), ino, &len);
 242
 243        /*
 244         * Print the dentry name for named mappings, and a
 245         * special [heap] marker for the heap:
 246         */
 247        if (file) {
 248                pad_len_spaces(m, len);
 249                seq_path(m, &file->f_path, "\n");
 250        } else {
 251                const char *name = arch_vma_name(vma);
 252                if (!name) {
 253                        if (mm) {
 254                                if (vma->vm_start <= mm->brk &&
 255                                                vma->vm_end >= mm->start_brk) {
 256                                        name = "[heap]";
 257                                } else if (vma->vm_start <= mm->start_stack &&
 258                                           vma->vm_end >= mm->start_stack) {
 259                                        name = "[stack]";
 260                                }
 261                        } else {
 262                                name = "[vdso]";
 263                        }
 264                }
 265                if (name) {
 266                        pad_len_spaces(m, len);
 267                        seq_puts(m, name);
 268                }
 269        }
 270        seq_putc(m, '\n');
 271}
 272
 273static int show_map(struct seq_file *m, void *v)
 274{
 275        struct vm_area_struct *vma = v;
 276        struct proc_maps_private *priv = m->private;
 277        struct task_struct *task = priv->task;
 278
 279        show_map_vma(m, vma);
 280
 281        if (m->count < m->size)  /* vma is copied successfully */
 282                m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
 283        return 0;
 284}
 285
 286static const struct seq_operations proc_pid_maps_op = {
 287        .start  = m_start,
 288        .next   = m_next,
 289        .stop   = m_stop,
 290        .show   = show_map
 291};
 292
 293static int maps_open(struct inode *inode, struct file *file)
 294{
 295        return do_maps_open(inode, file, &proc_pid_maps_op);
 296}
 297
 298const struct file_operations proc_maps_operations = {
 299        .open           = maps_open,
 300        .read           = seq_read,
 301        .llseek         = seq_lseek,
 302        .release        = seq_release_private,
 303};
 304
 305/*
 306 * Proportional Set Size(PSS): my share of RSS.
 307 *
 308 * PSS of a process is the count of pages it has in memory, where each
 309 * page is divided by the number of processes sharing it.  So if a
 310 * process has 1000 pages all to itself, and 1000 shared with one other
 311 * process, its PSS will be 1500.
 312 *
 313 * To keep (accumulated) division errors low, we adopt a 64bit
 314 * fixed-point pss counter to minimize division errors. So (pss >>
 315 * PSS_SHIFT) would be the real byte count.
 316 *
 317 * A shift of 12 before division means (assuming 4K page size):
 318 *      - 1M 3-user-pages add up to 8KB errors;
 319 *      - supports mapcount up to 2^24, or 16M;
 320 *      - supports PSS up to 2^52 bytes, or 4PB.
 321 */
 322#define PSS_SHIFT 12
 323
 324#ifdef CONFIG_PROC_PAGE_MONITOR
 325struct mem_size_stats {
 326        struct vm_area_struct *vma;
 327        unsigned long resident;
 328        unsigned long shared_clean;
 329        unsigned long shared_dirty;
 330        unsigned long private_clean;
 331        unsigned long private_dirty;
 332        unsigned long referenced;
 333        unsigned long anonymous;
 334        unsigned long swap;
 335        u64 pss;
 336};
 337
 338static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 339                           struct mm_walk *walk)
 340{
 341        struct mem_size_stats *mss = walk->private;
 342        struct vm_area_struct *vma = mss->vma;
 343        pte_t *pte, ptent;
 344        spinlock_t *ptl;
 345        struct page *page;
 346        int mapcount;
 347
 348        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 349        for (; addr != end; pte++, addr += PAGE_SIZE) {
 350                ptent = *pte;
 351
 352                if (is_swap_pte(ptent)) {
 353                        mss->swap += PAGE_SIZE;
 354                        continue;
 355                }
 356
 357                if (!pte_present(ptent))
 358                        continue;
 359
 360                page = vm_normal_page(vma, addr, ptent);
 361                if (!page)
 362                        continue;
 363
 364                if (PageAnon(page))
 365                        mss->anonymous += PAGE_SIZE;
 366
 367                mss->resident += PAGE_SIZE;
 368                /* Accumulate the size in pages that have been accessed. */
 369                if (pte_young(ptent) || PageReferenced(page))
 370                        mss->referenced += PAGE_SIZE;
 371                mapcount = page_mapcount(page);
 372                if (mapcount >= 2) {
 373                        if (pte_dirty(ptent) || PageDirty(page))
 374                                mss->shared_dirty += PAGE_SIZE;
 375                        else
 376                                mss->shared_clean += PAGE_SIZE;
 377                        mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
 378                } else {
 379                        if (pte_dirty(ptent) || PageDirty(page))
 380                                mss->private_dirty += PAGE_SIZE;
 381                        else
 382                                mss->private_clean += PAGE_SIZE;
 383                        mss->pss += (PAGE_SIZE << PSS_SHIFT);
 384                }
 385        }
 386        pte_unmap_unlock(pte - 1, ptl);
 387        cond_resched();
 388        return 0;
 389}
 390
 391static int show_smap(struct seq_file *m, void *v)
 392{
 393        struct proc_maps_private *priv = m->private;
 394        struct task_struct *task = priv->task;
 395        struct vm_area_struct *vma = v;
 396        struct mem_size_stats mss;
 397        struct mm_walk smaps_walk = {
 398                .pmd_entry = smaps_pte_range,
 399                .mm = vma->vm_mm,
 400                .private = &mss,
 401        };
 402
 403        memset(&mss, 0, sizeof mss);
 404        mss.vma = vma;
 405        /* mmap_sem is held in m_start */
 406        if (vma->vm_mm && !is_vm_hugetlb_page(vma))
 407                walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
 408
 409        show_map_vma(m, vma);
 410
 411        seq_printf(m,
 412                   "Size:           %8lu kB\n"
 413                   "Rss:            %8lu kB\n"
 414                   "Pss:            %8lu kB\n"
 415                   "Shared_Clean:   %8lu kB\n"
 416                   "Shared_Dirty:   %8lu kB\n"
 417                   "Private_Clean:  %8lu kB\n"
 418                   "Private_Dirty:  %8lu kB\n"
 419                   "Referenced:     %8lu kB\n"
 420                   "Anonymous:      %8lu kB\n"
 421                   "Swap:           %8lu kB\n"
 422                   "KernelPageSize: %8lu kB\n"
 423                   "MMUPageSize:    %8lu kB\n"
 424                   "Locked:         %8lu kB\n",
 425                   (vma->vm_end - vma->vm_start) >> 10,
 426                   mss.resident >> 10,
 427                   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 428                   mss.shared_clean  >> 10,
 429                   mss.shared_dirty  >> 10,
 430                   mss.private_clean >> 10,
 431                   mss.private_dirty >> 10,
 432                   mss.referenced >> 10,
 433                   mss.anonymous >> 10,
 434                   mss.swap >> 10,
 435                   vma_kernel_pagesize(vma) >> 10,
 436                   vma_mmu_pagesize(vma) >> 10,
 437                   (vma->vm_flags & VM_LOCKED) ?
 438                        (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 439
 440        if (m->count < m->size)  /* vma is copied successfully */
 441                m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
 442        return 0;
 443}
 444
 445static const struct seq_operations proc_pid_smaps_op = {
 446        .start  = m_start,
 447        .next   = m_next,
 448        .stop   = m_stop,
 449        .show   = show_smap
 450};
 451
 452static int smaps_open(struct inode *inode, struct file *file)
 453{
 454        return do_maps_open(inode, file, &proc_pid_smaps_op);
 455}
 456
 457const struct file_operations proc_smaps_operations = {
 458        .open           = smaps_open,
 459        .read           = seq_read,
 460        .llseek         = seq_lseek,
 461        .release        = seq_release_private,
 462};
 463
 464static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 465                                unsigned long end, struct mm_walk *walk)
 466{
 467        struct vm_area_struct *vma = walk->private;
 468        pte_t *pte, ptent;
 469        spinlock_t *ptl;
 470        struct page *page;
 471
 472        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 473        for (; addr != end; pte++, addr += PAGE_SIZE) {
 474                ptent = *pte;
 475                if (!pte_present(ptent))
 476                        continue;
 477
 478                page = vm_normal_page(vma, addr, ptent);
 479                if (!page)
 480                        continue;
 481
 482                /* Clear accessed and referenced bits. */
 483                ptep_test_and_clear_young(vma, addr, pte);
 484                ClearPageReferenced(page);
 485        }
 486        pte_unmap_unlock(pte - 1, ptl);
 487        cond_resched();
 488        return 0;
 489}
 490
 491#define CLEAR_REFS_ALL 1
 492#define CLEAR_REFS_ANON 2
 493#define CLEAR_REFS_MAPPED 3
 494
 495static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 496                                size_t count, loff_t *ppos)
 497{
 498        struct task_struct *task;
 499        char buffer[PROC_NUMBUF];
 500        struct mm_struct *mm;
 501        struct vm_area_struct *vma;
 502        long type;
 503
 504        memset(buffer, 0, sizeof(buffer));
 505        if (count > sizeof(buffer) - 1)
 506                count = sizeof(buffer) - 1;
 507        if (copy_from_user(buffer, buf, count))
 508                return -EFAULT;
 509        if (strict_strtol(strstrip(buffer), 10, &type))
 510                return -EINVAL;
 511        if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
 512                return -EINVAL;
 513        task = get_proc_task(file->f_path.dentry->d_inode);
 514        if (!task)
 515                return -ESRCH;
 516        mm = get_task_mm(task);
 517        if (mm) {
 518                struct mm_walk clear_refs_walk = {
 519                        .pmd_entry = clear_refs_pte_range,
 520                        .mm = mm,
 521                };
 522                down_read(&mm->mmap_sem);
 523                for (vma = mm->mmap; vma; vma = vma->vm_next) {
 524                        clear_refs_walk.private = vma;
 525                        if (is_vm_hugetlb_page(vma))
 526                                continue;
 527                        /*
 528                         * Writing 1 to /proc/pid/clear_refs affects all pages.
 529                         *
 530                         * Writing 2 to /proc/pid/clear_refs only affects
 531                         * Anonymous pages.
 532                         *
 533                         * Writing 3 to /proc/pid/clear_refs only affects file
 534                         * mapped pages.
 535                         */
 536                        if (type == CLEAR_REFS_ANON && vma->vm_file)
 537                                continue;
 538                        if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
 539                                continue;
 540                        walk_page_range(vma->vm_start, vma->vm_end,
 541                                        &clear_refs_walk);
 542                }
 543                flush_tlb_mm(mm);
 544                up_read(&mm->mmap_sem);
 545                mmput(mm);
 546        }
 547        put_task_struct(task);
 548
 549        return count;
 550}
 551
 552const struct file_operations proc_clear_refs_operations = {
 553        .write          = clear_refs_write,
 554        .llseek         = noop_llseek,
 555};
 556
 557struct pagemapread {
 558        int pos, len;
 559        u64 *buffer;
 560};
 561
 562#define PM_ENTRY_BYTES      sizeof(u64)
 563#define PM_STATUS_BITS      3
 564#define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
 565#define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
 566#define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
 567#define PM_PSHIFT_BITS      6
 568#define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
 569#define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
 570#define PM_PSHIFT(x)        (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
 571#define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
 572#define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
 573
 574#define PM_PRESENT          PM_STATUS(4LL)
 575#define PM_SWAP             PM_STATUS(2LL)
 576#define PM_NOT_PRESENT      PM_PSHIFT(PAGE_SHIFT)
 577#define PM_END_OF_BUFFER    1
 578
 579static int add_to_pagemap(unsigned long addr, u64 pfn,
 580                          struct pagemapread *pm)
 581{
 582        pm->buffer[pm->pos++] = pfn;
 583        if (pm->pos >= pm->len)
 584                return PM_END_OF_BUFFER;
 585        return 0;
 586}
 587
 588static int pagemap_pte_hole(unsigned long start, unsigned long end,
 589                                struct mm_walk *walk)
 590{
 591        struct pagemapread *pm = walk->private;
 592        unsigned long addr;
 593        int err = 0;
 594        for (addr = start; addr < end; addr += PAGE_SIZE) {
 595                err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
 596                if (err)
 597                        break;
 598        }
 599        return err;
 600}
 601
 602static u64 swap_pte_to_pagemap_entry(pte_t pte)
 603{
 604        swp_entry_t e = pte_to_swp_entry(pte);
 605        return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
 606}
 607
 608static u64 pte_to_pagemap_entry(pte_t pte)
 609{
 610        u64 pme = 0;
 611        if (is_swap_pte(pte))
 612                pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
 613                        | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
 614        else if (pte_present(pte))
 615                pme = PM_PFRAME(pte_pfn(pte))
 616                        | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
 617        return pme;
 618}
 619
 620static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 621                             struct mm_walk *walk)
 622{
 623        struct vm_area_struct *vma;
 624        struct pagemapread *pm = walk->private;
 625        pte_t *pte;
 626        int err = 0;
 627
 628        /* find the first VMA at or above 'addr' */
 629        vma = find_vma(walk->mm, addr);
 630        for (; addr != end; addr += PAGE_SIZE) {
 631                u64 pfn = PM_NOT_PRESENT;
 632
 633                /* check to see if we've left 'vma' behind
 634                 * and need a new, higher one */
 635                if (vma && (addr >= vma->vm_end))
 636                        vma = find_vma(walk->mm, addr);
 637
 638                /* check that 'vma' actually covers this address,
 639                 * and that it isn't a huge page vma */
 640                if (vma && (vma->vm_start <= addr) &&
 641                    !is_vm_hugetlb_page(vma)) {
 642                        pte = pte_offset_map(pmd, addr);
 643                        pfn = pte_to_pagemap_entry(*pte);
 644                        /* unmap before userspace copy */
 645                        pte_unmap(pte);
 646                }
 647                err = add_to_pagemap(addr, pfn, pm);
 648                if (err)
 649                        return err;
 650        }
 651
 652        cond_resched();
 653
 654        return err;
 655}
 656
 657#ifdef CONFIG_HUGETLB_PAGE
 658static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
 659{
 660        u64 pme = 0;
 661        if (pte_present(pte))
 662                pme = PM_PFRAME(pte_pfn(pte) + offset)
 663                        | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
 664        return pme;
 665}
 666
 667/* This function walks within one hugetlb entry in the single call */
 668static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
 669                                 unsigned long addr, unsigned long end,
 670                                 struct mm_walk *walk)
 671{
 672        struct pagemapread *pm = walk->private;
 673        int err = 0;
 674        u64 pfn;
 675
 676        for (; addr != end; addr += PAGE_SIZE) {
 677                int offset = (addr & ~hmask) >> PAGE_SHIFT;
 678                pfn = huge_pte_to_pagemap_entry(*pte, offset);
 679                err = add_to_pagemap(addr, pfn, pm);
 680                if (err)
 681                        return err;
 682        }
 683
 684        cond_resched();
 685
 686        return err;
 687}
 688#endif /* HUGETLB_PAGE */
 689
 690/*
 691 * /proc/pid/pagemap - an array mapping virtual pages to pfns
 692 *
 693 * For each page in the address space, this file contains one 64-bit entry
 694 * consisting of the following:
 695 *
 696 * Bits 0-55  page frame number (PFN) if present
 697 * Bits 0-4   swap type if swapped
 698 * Bits 5-55  swap offset if swapped
 699 * Bits 55-60 page shift (page size = 1<<page shift)
 700 * Bit  61    reserved for future use
 701 * Bit  62    page swapped
 702 * Bit  63    page present
 703 *
 704 * If the page is not present but in swap, then the PFN contains an
 705 * encoding of the swap file number and the page's offset into the
 706 * swap. Unmapped pages return a null PFN. This allows determining
 707 * precisely which pages are mapped (or in swap) and comparing mapped
 708 * pages between processes.
 709 *
 710 * Efficient users of this interface will use /proc/pid/maps to
 711 * determine which areas of memory are actually mapped and llseek to
 712 * skip over unmapped regions.
 713 */
 714#define PAGEMAP_WALK_SIZE       (PMD_SIZE)
 715#define PAGEMAP_WALK_MASK       (PMD_MASK)
 716static ssize_t pagemap_read(struct file *file, char __user *buf,
 717                            size_t count, loff_t *ppos)
 718{
 719        struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 720        struct mm_struct *mm;
 721        struct pagemapread pm;
 722        int ret = -ESRCH;
 723        struct mm_walk pagemap_walk = {};
 724        unsigned long src;
 725        unsigned long svpfn;
 726        unsigned long start_vaddr;
 727        unsigned long end_vaddr;
 728        int copied = 0;
 729
 730        if (!task)
 731                goto out;
 732
 733        ret = -EACCES;
 734        if (!ptrace_may_access(task, PTRACE_MODE_READ))
 735                goto out_task;
 736
 737        ret = -EINVAL;
 738        /* file position must be aligned */
 739        if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
 740                goto out_task;
 741
 742        ret = 0;
 743
 744        if (!count)
 745                goto out_task;
 746
 747        mm = get_task_mm(task);
 748        if (!mm)
 749                goto out_task;
 750
 751        pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
 752        pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
 753        ret = -ENOMEM;
 754        if (!pm.buffer)
 755                goto out_mm;
 756
 757        pagemap_walk.pmd_entry = pagemap_pte_range;
 758        pagemap_walk.pte_hole = pagemap_pte_hole;
 759#ifdef CONFIG_HUGETLB_PAGE
 760        pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
 761#endif
 762        pagemap_walk.mm = mm;
 763        pagemap_walk.private = &pm;
 764
 765        src = *ppos;
 766        svpfn = src / PM_ENTRY_BYTES;
 767        start_vaddr = svpfn << PAGE_SHIFT;
 768        end_vaddr = TASK_SIZE_OF(task);
 769
 770        /* watch out for wraparound */
 771        if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
 772                start_vaddr = end_vaddr;
 773
 774        /*
 775         * The odds are that this will stop walking way
 776         * before end_vaddr, because the length of the
 777         * user buffer is tracked in "pm", and the walk
 778         * will stop when we hit the end of the buffer.
 779         */
 780        ret = 0;
 781        while (count && (start_vaddr < end_vaddr)) {
 782                int len;
 783                unsigned long end;
 784
 785                pm.pos = 0;
 786                end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
 787                /* overflow ? */
 788                if (end < start_vaddr || end > end_vaddr)
 789                        end = end_vaddr;
 790                down_read(&mm->mmap_sem);
 791                ret = walk_page_range(start_vaddr, end, &pagemap_walk);
 792                up_read(&mm->mmap_sem);
 793                start_vaddr = end;
 794
 795                len = min(count, PM_ENTRY_BYTES * pm.pos);
 796                if (copy_to_user(buf, pm.buffer, len)) {
 797                        ret = -EFAULT;
 798                        goto out_free;
 799                }
 800                copied += len;
 801                buf += len;
 802                count -= len;
 803        }
 804        *ppos += copied;
 805        if (!ret || ret == PM_END_OF_BUFFER)
 806                ret = copied;
 807
 808out_free:
 809        kfree(pm.buffer);
 810out_mm:
 811        mmput(mm);
 812out_task:
 813        put_task_struct(task);
 814out:
 815        return ret;
 816}
 817
 818const struct file_operations proc_pagemap_operations = {
 819        .llseek         = mem_lseek, /* borrow this */
 820        .read           = pagemap_read,
 821};
 822#endif /* CONFIG_PROC_PAGE_MONITOR */
 823
 824#ifdef CONFIG_NUMA
 825extern int show_numa_map(struct seq_file *m, void *v);
 826
 827static const struct seq_operations proc_pid_numa_maps_op = {
 828        .start  = m_start,
 829        .next   = m_next,
 830        .stop   = m_stop,
 831        .show   = show_numa_map,
 832};
 833
 834static int numa_maps_open(struct inode *inode, struct file *file)
 835{
 836        return do_maps_open(inode, file, &proc_pid_numa_maps_op);
 837}
 838
 839const struct file_operations proc_numa_maps_operations = {
 840        .open           = numa_maps_open,
 841        .read           = seq_read,
 842        .llseek         = seq_lseek,
 843        .release        = seq_release_private,
 844};
 845#endif
 846