linux/fs/proc/task_mmu.c
<<
>>
Prefs
   1#include <linux/mm.h>
   2#include <linux/hugetlb.h>
   3#include <linux/huge_mm.h>
   4#include <linux/mount.h>
   5#include <linux/seq_file.h>
   6#include <linux/highmem.h>
   7#include <linux/ptrace.h>
   8#include <linux/slab.h>
   9#include <linux/pagemap.h>
  10#include <linux/mempolicy.h>
  11#include <linux/rmap.h>
  12#include <linux/swap.h>
  13#include <linux/swapops.h>
  14
  15#include <asm/elf.h>
  16#include <asm/uaccess.h>
  17#include <asm/tlbflush.h>
  18#include "internal.h"
  19
  20void task_mem(struct seq_file *m, struct mm_struct *mm)
  21{
  22        unsigned long data, text, lib, swap;
  23        unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  24
  25        /*
  26         * Note: to minimize their overhead, mm maintains hiwater_vm and
  27         * hiwater_rss only when about to *lower* total_vm or rss.  Any
  28         * collector of these hiwater stats must therefore get total_vm
  29         * and rss too, which will usually be the higher.  Barriers? not
  30         * worth the effort, such snapshots can always be inconsistent.
  31         */
  32        hiwater_vm = total_vm = mm->total_vm;
  33        if (hiwater_vm < mm->hiwater_vm)
  34                hiwater_vm = mm->hiwater_vm;
  35        hiwater_rss = total_rss = get_mm_rss(mm);
  36        if (hiwater_rss < mm->hiwater_rss)
  37                hiwater_rss = mm->hiwater_rss;
  38
  39        data = mm->total_vm - mm->shared_vm - mm->stack_vm;
  40        text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  41        lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  42        swap = get_mm_counter(mm, MM_SWAPENTS);
  43        seq_printf(m,
  44                "VmPeak:\t%8lu kB\n"
  45                "VmSize:\t%8lu kB\n"
  46                "VmLck:\t%8lu kB\n"
  47                "VmHWM:\t%8lu kB\n"
  48                "VmRSS:\t%8lu kB\n"
  49                "VmData:\t%8lu kB\n"
  50                "VmStk:\t%8lu kB\n"
  51                "VmExe:\t%8lu kB\n"
  52                "VmLib:\t%8lu kB\n"
  53                "VmPTE:\t%8lu kB\n"
  54                "VmSwap:\t%8lu kB\n",
  55                hiwater_vm << (PAGE_SHIFT-10),
  56                (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
  57                mm->locked_vm << (PAGE_SHIFT-10),
  58                hiwater_rss << (PAGE_SHIFT-10),
  59                total_rss << (PAGE_SHIFT-10),
  60                data << (PAGE_SHIFT-10),
  61                mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  62                (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
  63                swap << (PAGE_SHIFT-10));
  64}
  65
  66unsigned long task_vsize(struct mm_struct *mm)
  67{
  68        return PAGE_SIZE * mm->total_vm;
  69}
  70
  71unsigned long task_statm(struct mm_struct *mm,
  72                         unsigned long *shared, unsigned long *text,
  73                         unsigned long *data, unsigned long *resident)
  74{
  75        *shared = get_mm_counter(mm, MM_FILEPAGES);
  76        *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  77                                                                >> PAGE_SHIFT;
  78        *data = mm->total_vm - mm->shared_vm;
  79        *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
  80        return mm->total_vm;
  81}
  82
  83static void pad_len_spaces(struct seq_file *m, int len)
  84{
  85        len = 25 + sizeof(void*) * 6 - len;
  86        if (len < 1)
  87                len = 1;
  88        seq_printf(m, "%*c", len, ' ');
  89}
  90
  91static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
  92{
  93        if (vma && vma != priv->tail_vma) {
  94                struct mm_struct *mm = vma->vm_mm;
  95                up_read(&mm->mmap_sem);
  96                mmput(mm);
  97        }
  98}
  99
 100static void *m_start(struct seq_file *m, loff_t *pos)
 101{
 102        struct proc_maps_private *priv = m->private;
 103        unsigned long last_addr = m->version;
 104        struct mm_struct *mm;
 105        struct vm_area_struct *vma, *tail_vma = NULL;
 106        loff_t l = *pos;
 107
 108        /* Clear the per syscall fields in priv */
 109        priv->task = NULL;
 110        priv->tail_vma = NULL;
 111
 112        /*
 113         * We remember last_addr rather than next_addr to hit with
 114         * mmap_cache most of the time. We have zero last_addr at
 115         * the beginning and also after lseek. We will have -1 last_addr
 116         * after the end of the vmas.
 117         */
 118
 119        if (last_addr == -1UL)
 120                return NULL;
 121
 122        priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
 123        if (!priv->task)
 124                return ERR_PTR(-ESRCH);
 125
 126        mm = mm_for_maps(priv->task);
 127        if (!mm || IS_ERR(mm))
 128                return mm;
 129        down_read(&mm->mmap_sem);
 130
 131        tail_vma = get_gate_vma(priv->task->mm);
 132        priv->tail_vma = tail_vma;
 133
 134        /* Start with last addr hint */
 135        vma = find_vma(mm, last_addr);
 136        if (last_addr && vma) {
 137                vma = vma->vm_next;
 138                goto out;
 139        }
 140
 141        /*
 142         * Check the vma index is within the range and do
 143         * sequential scan until m_index.
 144         */
 145        vma = NULL;
 146        if ((unsigned long)l < mm->map_count) {
 147                vma = mm->mmap;
 148                while (l-- && vma)
 149                        vma = vma->vm_next;
 150                goto out;
 151        }
 152
 153        if (l != mm->map_count)
 154                tail_vma = NULL; /* After gate vma */
 155
 156out:
 157        if (vma)
 158                return vma;
 159
 160        /* End of vmas has been reached */
 161        m->version = (tail_vma != NULL)? 0: -1UL;
 162        up_read(&mm->mmap_sem);
 163        mmput(mm);
 164        return tail_vma;
 165}
 166
 167static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 168{
 169        struct proc_maps_private *priv = m->private;
 170        struct vm_area_struct *vma = v;
 171        struct vm_area_struct *tail_vma = priv->tail_vma;
 172
 173        (*pos)++;
 174        if (vma && (vma != tail_vma) && vma->vm_next)
 175                return vma->vm_next;
 176        vma_stop(priv, vma);
 177        return (vma != tail_vma)? tail_vma: NULL;
 178}
 179
 180static void m_stop(struct seq_file *m, void *v)
 181{
 182        struct proc_maps_private *priv = m->private;
 183        struct vm_area_struct *vma = v;
 184
 185        if (!IS_ERR(vma))
 186                vma_stop(priv, vma);
 187        if (priv->task)
 188                put_task_struct(priv->task);
 189}
 190
 191static int do_maps_open(struct inode *inode, struct file *file,
 192                        const struct seq_operations *ops)
 193{
 194        struct proc_maps_private *priv;
 195        int ret = -ENOMEM;
 196        priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 197        if (priv) {
 198                priv->pid = proc_pid(inode);
 199                ret = seq_open(file, ops);
 200                if (!ret) {
 201                        struct seq_file *m = file->private_data;
 202                        m->private = priv;
 203                } else {
 204                        kfree(priv);
 205                }
 206        }
 207        return ret;
 208}
 209
 210static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 211{
 212        struct mm_struct *mm = vma->vm_mm;
 213        struct file *file = vma->vm_file;
 214        int flags = vma->vm_flags;
 215        unsigned long ino = 0;
 216        unsigned long long pgoff = 0;
 217        unsigned long start, end;
 218        dev_t dev = 0;
 219        int len;
 220
 221        if (file) {
 222                struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
 223                dev = inode->i_sb->s_dev;
 224                ino = inode->i_ino;
 225                pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 226        }
 227
 228        /* We don't show the stack guard page in /proc/maps */
 229        start = vma->vm_start;
 230        if (stack_guard_page_start(vma, start))
 231                start += PAGE_SIZE;
 232        end = vma->vm_end;
 233        if (stack_guard_page_end(vma, end))
 234                end -= PAGE_SIZE;
 235
 236        seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
 237                        start,
 238                        end,
 239                        flags & VM_READ ? 'r' : '-',
 240                        flags & VM_WRITE ? 'w' : '-',
 241                        flags & VM_EXEC ? 'x' : '-',
 242                        flags & VM_MAYSHARE ? 's' : 'p',
 243                        pgoff,
 244                        MAJOR(dev), MINOR(dev), ino, &len);
 245
 246        /*
 247         * Print the dentry name for named mappings, and a
 248         * special [heap] marker for the heap:
 249         */
 250        if (file) {
 251                pad_len_spaces(m, len);
 252                seq_path(m, &file->f_path, "\n");
 253        } else {
 254                const char *name = arch_vma_name(vma);
 255                if (!name) {
 256                        if (mm) {
 257                                if (vma->vm_start <= mm->brk &&
 258                                                vma->vm_end >= mm->start_brk) {
 259                                        name = "[heap]";
 260                                } else if (vma->vm_start <= mm->start_stack &&
 261                                           vma->vm_end >= mm->start_stack) {
 262                                        name = "[stack]";
 263                                }
 264                        } else {
 265                                name = "[vdso]";
 266                        }
 267                }
 268                if (name) {
 269                        pad_len_spaces(m, len);
 270                        seq_puts(m, name);
 271                }
 272        }
 273        seq_putc(m, '\n');
 274}
 275
 276static int show_map(struct seq_file *m, void *v)
 277{
 278        struct vm_area_struct *vma = v;
 279        struct proc_maps_private *priv = m->private;
 280        struct task_struct *task = priv->task;
 281
 282        show_map_vma(m, vma);
 283
 284        if (m->count < m->size)  /* vma is copied successfully */
 285                m->version = (vma != get_gate_vma(task->mm))
 286                        ? vma->vm_start : 0;
 287        return 0;
 288}
 289
 290static const struct seq_operations proc_pid_maps_op = {
 291        .start  = m_start,
 292        .next   = m_next,
 293        .stop   = m_stop,
 294        .show   = show_map
 295};
 296
 297static int maps_open(struct inode *inode, struct file *file)
 298{
 299        return do_maps_open(inode, file, &proc_pid_maps_op);
 300}
 301
 302const struct file_operations proc_maps_operations = {
 303        .open           = maps_open,
 304        .read           = seq_read,
 305        .llseek         = seq_lseek,
 306        .release        = seq_release_private,
 307};
 308
 309/*
 310 * Proportional Set Size(PSS): my share of RSS.
 311 *
 312 * PSS of a process is the count of pages it has in memory, where each
 313 * page is divided by the number of processes sharing it.  So if a
 314 * process has 1000 pages all to itself, and 1000 shared with one other
 315 * process, its PSS will be 1500.
 316 *
 317 * To keep (accumulated) division errors low, we adopt a 64bit
 318 * fixed-point pss counter to minimize division errors. So (pss >>
 319 * PSS_SHIFT) would be the real byte count.
 320 *
 321 * A shift of 12 before division means (assuming 4K page size):
 322 *      - 1M 3-user-pages add up to 8KB errors;
 323 *      - supports mapcount up to 2^24, or 16M;
 324 *      - supports PSS up to 2^52 bytes, or 4PB.
 325 */
 326#define PSS_SHIFT 12
 327
 328#ifdef CONFIG_PROC_PAGE_MONITOR
 329struct mem_size_stats {
 330        struct vm_area_struct *vma;
 331        unsigned long resident;
 332        unsigned long shared_clean;
 333        unsigned long shared_dirty;
 334        unsigned long private_clean;
 335        unsigned long private_dirty;
 336        unsigned long referenced;
 337        unsigned long anonymous;
 338        unsigned long anonymous_thp;
 339        unsigned long swap;
 340        u64 pss;
 341};
 342
 343
 344static void smaps_pte_entry(pte_t ptent, unsigned long addr,
 345                unsigned long ptent_size, struct mm_walk *walk)
 346{
 347        struct mem_size_stats *mss = walk->private;
 348        struct vm_area_struct *vma = mss->vma;
 349        struct page *page;
 350        int mapcount;
 351
 352        if (is_swap_pte(ptent)) {
 353                mss->swap += ptent_size;
 354                return;
 355        }
 356
 357        if (!pte_present(ptent))
 358                return;
 359
 360        page = vm_normal_page(vma, addr, ptent);
 361        if (!page)
 362                return;
 363
 364        if (PageAnon(page))
 365                mss->anonymous += ptent_size;
 366
 367        mss->resident += ptent_size;
 368        /* Accumulate the size in pages that have been accessed. */
 369        if (pte_young(ptent) || PageReferenced(page))
 370                mss->referenced += ptent_size;
 371        mapcount = page_mapcount(page);
 372        if (mapcount >= 2) {
 373                if (pte_dirty(ptent) || PageDirty(page))
 374                        mss->shared_dirty += ptent_size;
 375                else
 376                        mss->shared_clean += ptent_size;
 377                mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
 378        } else {
 379                if (pte_dirty(ptent) || PageDirty(page))
 380                        mss->private_dirty += ptent_size;
 381                else
 382                        mss->private_clean += ptent_size;
 383                mss->pss += (ptent_size << PSS_SHIFT);
 384        }
 385}
 386
 387static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 388                           struct mm_walk *walk)
 389{
 390        struct mem_size_stats *mss = walk->private;
 391        struct vm_area_struct *vma = mss->vma;
 392        pte_t *pte;
 393        spinlock_t *ptl;
 394
 395        spin_lock(&walk->mm->page_table_lock);
 396        if (pmd_trans_huge(*pmd)) {
 397                if (pmd_trans_splitting(*pmd)) {
 398                        spin_unlock(&walk->mm->page_table_lock);
 399                        wait_split_huge_page(vma->anon_vma, pmd);
 400                } else {
 401                        smaps_pte_entry(*(pte_t *)pmd, addr,
 402                                        HPAGE_PMD_SIZE, walk);
 403                        spin_unlock(&walk->mm->page_table_lock);
 404                        mss->anonymous_thp += HPAGE_PMD_SIZE;
 405                        return 0;
 406                }
 407        } else {
 408                spin_unlock(&walk->mm->page_table_lock);
 409        }
 410        /*
 411         * The mmap_sem held all the way back in m_start() is what
 412         * keeps khugepaged out of here and from collapsing things
 413         * in here.
 414         */
 415        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 416        for (; addr != end; pte++, addr += PAGE_SIZE)
 417                smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
 418        pte_unmap_unlock(pte - 1, ptl);
 419        cond_resched();
 420        return 0;
 421}
 422
 423static int show_smap(struct seq_file *m, void *v)
 424{
 425        struct proc_maps_private *priv = m->private;
 426        struct task_struct *task = priv->task;
 427        struct vm_area_struct *vma = v;
 428        struct mem_size_stats mss;
 429        struct mm_walk smaps_walk = {
 430                .pmd_entry = smaps_pte_range,
 431                .mm = vma->vm_mm,
 432                .private = &mss,
 433        };
 434
 435        memset(&mss, 0, sizeof mss);
 436        mss.vma = vma;
 437        /* mmap_sem is held in m_start */
 438        if (vma->vm_mm && !is_vm_hugetlb_page(vma))
 439                walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
 440
 441        show_map_vma(m, vma);
 442
 443        seq_printf(m,
 444                   "Size:           %8lu kB\n"
 445                   "Rss:            %8lu kB\n"
 446                   "Pss:            %8lu kB\n"
 447                   "Shared_Clean:   %8lu kB\n"
 448                   "Shared_Dirty:   %8lu kB\n"
 449                   "Private_Clean:  %8lu kB\n"
 450                   "Private_Dirty:  %8lu kB\n"
 451                   "Referenced:     %8lu kB\n"
 452                   "Anonymous:      %8lu kB\n"
 453                   "AnonHugePages:  %8lu kB\n"
 454                   "Swap:           %8lu kB\n"
 455                   "KernelPageSize: %8lu kB\n"
 456                   "MMUPageSize:    %8lu kB\n"
 457                   "Locked:         %8lu kB\n",
 458                   (vma->vm_end - vma->vm_start) >> 10,
 459                   mss.resident >> 10,
 460                   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 461                   mss.shared_clean  >> 10,
 462                   mss.shared_dirty  >> 10,
 463                   mss.private_clean >> 10,
 464                   mss.private_dirty >> 10,
 465                   mss.referenced >> 10,
 466                   mss.anonymous >> 10,
 467                   mss.anonymous_thp >> 10,
 468                   mss.swap >> 10,
 469                   vma_kernel_pagesize(vma) >> 10,
 470                   vma_mmu_pagesize(vma) >> 10,
 471                   (vma->vm_flags & VM_LOCKED) ?
 472                        (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 473
 474        if (m->count < m->size)  /* vma is copied successfully */
 475                m->version = (vma != get_gate_vma(task->mm))
 476                        ? vma->vm_start : 0;
 477        return 0;
 478}
 479
 480static const struct seq_operations proc_pid_smaps_op = {
 481        .start  = m_start,
 482        .next   = m_next,
 483        .stop   = m_stop,
 484        .show   = show_smap
 485};
 486
 487static int smaps_open(struct inode *inode, struct file *file)
 488{
 489        return do_maps_open(inode, file, &proc_pid_smaps_op);
 490}
 491
 492const struct file_operations proc_smaps_operations = {
 493        .open           = smaps_open,
 494        .read           = seq_read,
 495        .llseek         = seq_lseek,
 496        .release        = seq_release_private,
 497};
 498
 499static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 500                                unsigned long end, struct mm_walk *walk)
 501{
 502        struct vm_area_struct *vma = walk->private;
 503        pte_t *pte, ptent;
 504        spinlock_t *ptl;
 505        struct page *page;
 506
 507        split_huge_page_pmd(walk->mm, pmd);
 508
 509        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 510        for (; addr != end; pte++, addr += PAGE_SIZE) {
 511                ptent = *pte;
 512                if (!pte_present(ptent))
 513                        continue;
 514
 515                page = vm_normal_page(vma, addr, ptent);
 516                if (!page)
 517                        continue;
 518
 519                /* Clear accessed and referenced bits. */
 520                ptep_test_and_clear_young(vma, addr, pte);
 521                ClearPageReferenced(page);
 522        }
 523        pte_unmap_unlock(pte - 1, ptl);
 524        cond_resched();
 525        return 0;
 526}
 527
 528#define CLEAR_REFS_ALL 1
 529#define CLEAR_REFS_ANON 2
 530#define CLEAR_REFS_MAPPED 3
 531
 532static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 533                                size_t count, loff_t *ppos)
 534{
 535        struct task_struct *task;
 536        char buffer[PROC_NUMBUF];
 537        struct mm_struct *mm;
 538        struct vm_area_struct *vma;
 539        long type;
 540
 541        memset(buffer, 0, sizeof(buffer));
 542        if (count > sizeof(buffer) - 1)
 543                count = sizeof(buffer) - 1;
 544        if (copy_from_user(buffer, buf, count))
 545                return -EFAULT;
 546        if (strict_strtol(strstrip(buffer), 10, &type))
 547                return -EINVAL;
 548        if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
 549                return -EINVAL;
 550        task = get_proc_task(file->f_path.dentry->d_inode);
 551        if (!task)
 552                return -ESRCH;
 553        mm = get_task_mm(task);
 554        if (mm) {
 555                struct mm_walk clear_refs_walk = {
 556                        .pmd_entry = clear_refs_pte_range,
 557                        .mm = mm,
 558                };
 559                down_read(&mm->mmap_sem);
 560                for (vma = mm->mmap; vma; vma = vma->vm_next) {
 561                        clear_refs_walk.private = vma;
 562                        if (is_vm_hugetlb_page(vma))
 563                                continue;
 564                        /*
 565                         * Writing 1 to /proc/pid/clear_refs affects all pages.
 566                         *
 567                         * Writing 2 to /proc/pid/clear_refs only affects
 568                         * Anonymous pages.
 569                         *
 570                         * Writing 3 to /proc/pid/clear_refs only affects file
 571                         * mapped pages.
 572                         */
 573                        if (type == CLEAR_REFS_ANON && vma->vm_file)
 574                                continue;
 575                        if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
 576                                continue;
 577                        walk_page_range(vma->vm_start, vma->vm_end,
 578                                        &clear_refs_walk);
 579                }
 580                flush_tlb_mm(mm);
 581                up_read(&mm->mmap_sem);
 582                mmput(mm);
 583        }
 584        put_task_struct(task);
 585
 586        return count;
 587}
 588
 589const struct file_operations proc_clear_refs_operations = {
 590        .write          = clear_refs_write,
 591        .llseek         = noop_llseek,
 592};
 593
 594struct pagemapread {
 595        int pos, len;
 596        u64 *buffer;
 597};
 598
 599#define PM_ENTRY_BYTES      sizeof(u64)
 600#define PM_STATUS_BITS      3
 601#define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
 602#define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
 603#define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
 604#define PM_PSHIFT_BITS      6
 605#define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
 606#define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
 607#define PM_PSHIFT(x)        (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
 608#define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
 609#define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
 610
 611#define PM_PRESENT          PM_STATUS(4LL)
 612#define PM_SWAP             PM_STATUS(2LL)
 613#define PM_NOT_PRESENT      PM_PSHIFT(PAGE_SHIFT)
 614#define PM_END_OF_BUFFER    1
 615
 616static int add_to_pagemap(unsigned long addr, u64 pfn,
 617                          struct pagemapread *pm)
 618{
 619        pm->buffer[pm->pos++] = pfn;
 620        if (pm->pos >= pm->len)
 621                return PM_END_OF_BUFFER;
 622        return 0;
 623}
 624
 625static int pagemap_pte_hole(unsigned long start, unsigned long end,
 626                                struct mm_walk *walk)
 627{
 628        struct pagemapread *pm = walk->private;
 629        unsigned long addr;
 630        int err = 0;
 631        for (addr = start; addr < end; addr += PAGE_SIZE) {
 632                err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
 633                if (err)
 634                        break;
 635        }
 636        return err;
 637}
 638
 639static u64 swap_pte_to_pagemap_entry(pte_t pte)
 640{
 641        swp_entry_t e = pte_to_swp_entry(pte);
 642        return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
 643}
 644
 645static u64 pte_to_pagemap_entry(pte_t pte)
 646{
 647        u64 pme = 0;
 648        if (is_swap_pte(pte))
 649                pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
 650                        | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
 651        else if (pte_present(pte))
 652                pme = PM_PFRAME(pte_pfn(pte))
 653                        | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
 654        return pme;
 655}
 656
 657static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 658                             struct mm_walk *walk)
 659{
 660        struct vm_area_struct *vma;
 661        struct pagemapread *pm = walk->private;
 662        pte_t *pte;
 663        int err = 0;
 664
 665        split_huge_page_pmd(walk->mm, pmd);
 666
 667        /* find the first VMA at or above 'addr' */
 668        vma = find_vma(walk->mm, addr);
 669        for (; addr != end; addr += PAGE_SIZE) {
 670                u64 pfn = PM_NOT_PRESENT;
 671
 672                /* check to see if we've left 'vma' behind
 673                 * and need a new, higher one */
 674                if (vma && (addr >= vma->vm_end))
 675                        vma = find_vma(walk->mm, addr);
 676
 677                /* check that 'vma' actually covers this address,
 678                 * and that it isn't a huge page vma */
 679                if (vma && (vma->vm_start <= addr) &&
 680                    !is_vm_hugetlb_page(vma)) {
 681                        pte = pte_offset_map(pmd, addr);
 682                        pfn = pte_to_pagemap_entry(*pte);
 683                        /* unmap before userspace copy */
 684                        pte_unmap(pte);
 685                }
 686                err = add_to_pagemap(addr, pfn, pm);
 687                if (err)
 688                        return err;
 689        }
 690
 691        cond_resched();
 692
 693        return err;
 694}
 695
 696#ifdef CONFIG_HUGETLB_PAGE
 697static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
 698{
 699        u64 pme = 0;
 700        if (pte_present(pte))
 701                pme = PM_PFRAME(pte_pfn(pte) + offset)
 702                        | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
 703        return pme;
 704}
 705
 706/* This function walks within one hugetlb entry in the single call */
 707static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
 708                                 unsigned long addr, unsigned long end,
 709                                 struct mm_walk *walk)
 710{
 711        struct pagemapread *pm = walk->private;
 712        int err = 0;
 713        u64 pfn;
 714
 715        for (; addr != end; addr += PAGE_SIZE) {
 716                int offset = (addr & ~hmask) >> PAGE_SHIFT;
 717                pfn = huge_pte_to_pagemap_entry(*pte, offset);
 718                err = add_to_pagemap(addr, pfn, pm);
 719                if (err)
 720                        return err;
 721        }
 722
 723        cond_resched();
 724
 725        return err;
 726}
 727#endif /* HUGETLB_PAGE */
 728
 729/*
 730 * /proc/pid/pagemap - an array mapping virtual pages to pfns
 731 *
 732 * For each page in the address space, this file contains one 64-bit entry
 733 * consisting of the following:
 734 *
 735 * Bits 0-55  page frame number (PFN) if present
 736 * Bits 0-4   swap type if swapped
 737 * Bits 5-55  swap offset if swapped
 738 * Bits 55-60 page shift (page size = 1<<page shift)
 739 * Bit  61    reserved for future use
 740 * Bit  62    page swapped
 741 * Bit  63    page present
 742 *
 743 * If the page is not present but in swap, then the PFN contains an
 744 * encoding of the swap file number and the page's offset into the
 745 * swap. Unmapped pages return a null PFN. This allows determining
 746 * precisely which pages are mapped (or in swap) and comparing mapped
 747 * pages between processes.
 748 *
 749 * Efficient users of this interface will use /proc/pid/maps to
 750 * determine which areas of memory are actually mapped and llseek to
 751 * skip over unmapped regions.
 752 */
 753#define PAGEMAP_WALK_SIZE       (PMD_SIZE)
 754#define PAGEMAP_WALK_MASK       (PMD_MASK)
 755static ssize_t pagemap_read(struct file *file, char __user *buf,
 756                            size_t count, loff_t *ppos)
 757{
 758        struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 759        struct mm_struct *mm;
 760        struct pagemapread pm;
 761        int ret = -ESRCH;
 762        struct mm_walk pagemap_walk = {};
 763        unsigned long src;
 764        unsigned long svpfn;
 765        unsigned long start_vaddr;
 766        unsigned long end_vaddr;
 767        int copied = 0;
 768
 769        if (!task)
 770                goto out;
 771
 772        mm = mm_for_maps(task);
 773        ret = PTR_ERR(mm);
 774        if (!mm || IS_ERR(mm))
 775                goto out_task;
 776
 777        ret = -EINVAL;
 778        /* file position must be aligned */
 779        if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
 780                goto out_task;
 781
 782        ret = 0;
 783
 784        if (!count)
 785                goto out_task;
 786
 787        pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
 788        pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
 789        ret = -ENOMEM;
 790        if (!pm.buffer)
 791                goto out_mm;
 792
 793        pagemap_walk.pmd_entry = pagemap_pte_range;
 794        pagemap_walk.pte_hole = pagemap_pte_hole;
 795#ifdef CONFIG_HUGETLB_PAGE
 796        pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
 797#endif
 798        pagemap_walk.mm = mm;
 799        pagemap_walk.private = &pm;
 800
 801        src = *ppos;
 802        svpfn = src / PM_ENTRY_BYTES;
 803        start_vaddr = svpfn << PAGE_SHIFT;
 804        end_vaddr = TASK_SIZE_OF(task);
 805
 806        /* watch out for wraparound */
 807        if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
 808                start_vaddr = end_vaddr;
 809
 810        /*
 811         * The odds are that this will stop walking way
 812         * before end_vaddr, because the length of the
 813         * user buffer is tracked in "pm", and the walk
 814         * will stop when we hit the end of the buffer.
 815         */
 816        ret = 0;
 817        while (count && (start_vaddr < end_vaddr)) {
 818                int len;
 819                unsigned long end;
 820
 821                pm.pos = 0;
 822                end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
 823                /* overflow ? */
 824                if (end < start_vaddr || end > end_vaddr)
 825                        end = end_vaddr;
 826                down_read(&mm->mmap_sem);
 827                ret = walk_page_range(start_vaddr, end, &pagemap_walk);
 828                up_read(&mm->mmap_sem);
 829                start_vaddr = end;
 830
 831                len = min(count, PM_ENTRY_BYTES * pm.pos);
 832                if (copy_to_user(buf, pm.buffer, len)) {
 833                        ret = -EFAULT;
 834                        goto out_free;
 835                }
 836                copied += len;
 837                buf += len;
 838                count -= len;
 839        }
 840        *ppos += copied;
 841        if (!ret || ret == PM_END_OF_BUFFER)
 842                ret = copied;
 843
 844out_free:
 845        kfree(pm.buffer);
 846out_mm:
 847        mmput(mm);
 848out_task:
 849        put_task_struct(task);
 850out:
 851        return ret;
 852}
 853
 854const struct file_operations proc_pagemap_operations = {
 855        .llseek         = mem_lseek, /* borrow this */
 856        .read           = pagemap_read,
 857};
 858#endif /* CONFIG_PROC_PAGE_MONITOR */
 859
 860#ifdef CONFIG_NUMA
 861extern int show_numa_map(struct seq_file *m, void *v);
 862
 863static const struct seq_operations proc_pid_numa_maps_op = {
 864        .start  = m_start,
 865        .next   = m_next,
 866        .stop   = m_stop,
 867        .show   = show_numa_map,
 868};
 869
 870static int numa_maps_open(struct inode *inode, struct file *file)
 871{
 872        return do_maps_open(inode, file, &proc_pid_numa_maps_op);
 873}
 874
 875const struct file_operations proc_numa_maps_operations = {
 876        .open           = numa_maps_open,
 877        .read           = seq_read,
 878        .llseek         = seq_lseek,
 879        .release        = seq_release_private,
 880};
 881#endif
 882