Merge branch 'llseek' of git://git.kernel.org/pub/scm/linux/kernel/git/arnd/bkl
[linux-2.6.git] / fs / proc / task_mmu.c
index 64e84ca..871e25e 100644 (file)
@@ -2,35 +2,63 @@
 #include <linux/hugetlb.h>
 #include <linux/mount.h>
 #include <linux/seq_file.h>
+#include <linux/highmem.h>
+#include <linux/ptrace.h>
+#include <linux/slab.h>
 #include <linux/pagemap.h>
 #include <linux/mempolicy.h>
+#include <linux/swap.h>
+#include <linux/swapops.h>
+
 #include <asm/elf.h>
 #include <asm/uaccess.h>
+#include <asm/tlbflush.h>
 #include "internal.h"
 
-char *task_mem(struct mm_struct *mm, char *buffer)
+void task_mem(struct seq_file *m, struct mm_struct *mm)
 {
-       unsigned long data, text, lib;
+       unsigned long data, text, lib, swap;
+       unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
+
+       /*
+        * Note: to minimize their overhead, mm maintains hiwater_vm and
+        * hiwater_rss only when about to *lower* total_vm or rss.  Any
+        * collector of these hiwater stats must therefore get total_vm
+        * and rss too, which will usually be the higher.  Barriers? not
+        * worth the effort, such snapshots can always be inconsistent.
+        */
+       hiwater_vm = total_vm = mm->total_vm;
+       if (hiwater_vm < mm->hiwater_vm)
+               hiwater_vm = mm->hiwater_vm;
+       hiwater_rss = total_rss = get_mm_rss(mm);
+       if (hiwater_rss < mm->hiwater_rss)
+               hiwater_rss = mm->hiwater_rss;
 
        data = mm->total_vm - mm->shared_vm - mm->stack_vm;
        text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
        lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
-       buffer += sprintf(buffer,
+       swap = get_mm_counter(mm, MM_SWAPENTS);
+       seq_printf(m,
+               "VmPeak:\t%8lu kB\n"
                "VmSize:\t%8lu kB\n"
                "VmLck:\t%8lu kB\n"
+               "VmHWM:\t%8lu kB\n"
                "VmRSS:\t%8lu kB\n"
                "VmData:\t%8lu kB\n"
                "VmStk:\t%8lu kB\n"
                "VmExe:\t%8lu kB\n"
                "VmLib:\t%8lu kB\n"
-               "VmPTE:\t%8lu kB\n",
-               (mm->total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
+               "VmPTE:\t%8lu kB\n"
+               "VmSwap:\t%8lu kB\n",
+               hiwater_vm << (PAGE_SHIFT-10),
+               (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
                mm->locked_vm << (PAGE_SHIFT-10),
-               get_mm_counter(mm, rss) << (PAGE_SHIFT-10),
+               hiwater_rss << (PAGE_SHIFT-10),
+               total_rss << (PAGE_SHIFT-10),
                data << (PAGE_SHIFT-10),
                mm->stack_vm << (PAGE_SHIFT-10), text, lib,
-               (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10);
-       return buffer;
+               (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
+               swap << (PAGE_SHIFT-10));
 }
 
 unsigned long task_vsize(struct mm_struct *mm)
@@ -41,46 +69,14 @@ unsigned long task_vsize(struct mm_struct *mm)
 int task_statm(struct mm_struct *mm, int *shared, int *text,
               int *data, int *resident)
 {
-       int rss = get_mm_counter(mm, rss);
-
-       *shared = rss - get_mm_counter(mm, anon_rss);
+       *shared = get_mm_counter(mm, MM_FILEPAGES);
        *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
                                                                >> PAGE_SHIFT;
        *data = mm->total_vm - mm->shared_vm;
-       *resident = rss;
+       *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
        return mm->total_vm;
 }
 
-int proc_exe_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
-{
-       struct vm_area_struct * vma;
-       int result = -ENOENT;
-       struct task_struct *task = proc_task(inode);
-       struct mm_struct * mm = get_task_mm(task);
-
-       if (!mm)
-               goto out;
-       down_read(&mm->mmap_sem);
-
-       vma = mm->mmap;
-       while (vma) {
-               if ((vma->vm_flags & VM_EXECUTABLE) && vma->vm_file)
-                       break;
-               vma = vma->vm_next;
-       }
-
-       if (vma) {
-               *mnt = mntget(vma->vm_file->f_vfsmnt);
-               *dentry = dget(vma->vm_file->f_dentry);
-               result = 0;
-       }
-
-       up_read(&mm->mmap_sem);
-       mmput(mm);
-out:
-       return result;
-}
-
 static void pad_len_spaces(struct seq_file *m, int len)
 {
        len = 25 + sizeof(void*) * 6 - len;
@@ -89,279 +85,748 @@ static void pad_len_spaces(struct seq_file *m, int len)
        seq_printf(m, "%*c", len, ' ');
 }
 
-static int show_map(struct seq_file *m, void *v)
+static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
 {
-       struct task_struct *task = m->private;
-       struct vm_area_struct *map = v;
-       struct mm_struct *mm = map->vm_mm;
-       struct file *file = map->vm_file;
-       int flags = map->vm_flags;
-       unsigned long ino = 0;
-       dev_t dev = 0;
-       int len;
-
-       if (file) {
-               struct inode *inode = map->vm_file->f_dentry->d_inode;
-               dev = inode->i_sb->s_dev;
-               ino = inode->i_ino;
-       }
-
-       seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
-                       map->vm_start,
-                       map->vm_end,
-                       flags & VM_READ ? 'r' : '-',
-                       flags & VM_WRITE ? 'w' : '-',
-                       flags & VM_EXEC ? 'x' : '-',
-                       flags & VM_MAYSHARE ? 's' : 'p',
-                       map->vm_pgoff << PAGE_SHIFT,
-                       MAJOR(dev), MINOR(dev), ino, &len);
-
-       /*
-        * Print the dentry name for named mappings, and a
-        * special [heap] marker for the heap:
-        */
-       if (map->vm_file) {
-               pad_len_spaces(m, len);
-               seq_path(m, file->f_vfsmnt, file->f_dentry, "");
-       } else {
-               if (mm) {
-                       if (map->vm_start <= mm->start_brk &&
-                                               map->vm_end >= mm->brk) {
-                               pad_len_spaces(m, len);
-                               seq_puts(m, "[heap]");
-                       } else {
-                               if (map->vm_start <= mm->start_stack &&
-                                       map->vm_end >= mm->start_stack) {
-
-                                       pad_len_spaces(m, len);
-                                       seq_puts(m, "[stack]");
-                               }
-                       }
-               } else {
-                       pad_len_spaces(m, len);
-                       seq_puts(m, "[vdso]");
-               }
+       if (vma && vma != priv->tail_vma) {
+               struct mm_struct *mm = vma->vm_mm;
+               up_read(&mm->mmap_sem);
+               mmput(mm);
        }
-       seq_putc(m, '\n');
-       if (m->count < m->size)  /* map is copied successfully */
-               m->version = (map != get_gate_vma(task))? map->vm_start: 0;
-       return 0;
 }
 
 static void *m_start(struct seq_file *m, loff_t *pos)
 {
-       struct task_struct *task = m->private;
+       struct proc_maps_private *priv = m->private;
        unsigned long last_addr = m->version;
        struct mm_struct *mm;
-       struct vm_area_struct *map, *tail_map;
+       struct vm_area_struct *vma, *tail_vma = NULL;
        loff_t l = *pos;
 
+       /* Clear the per syscall fields in priv */
+       priv->task = NULL;
+       priv->tail_vma = NULL;
+
        /*
         * We remember last_addr rather than next_addr to hit with
         * mmap_cache most of the time. We have zero last_addr at
-        * the begining and also after lseek. We will have -1 last_addr
-        * after the end of the maps.
+        * the beginning and also after lseek. We will have -1 last_addr
+        * after the end of the vmas.
         */
 
        if (last_addr == -1UL)
                return NULL;
 
-       mm = get_task_mm(task);
-       if (!mm)
+       priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
+       if (!priv->task)
                return NULL;
 
-       tail_map = get_gate_vma(task);
+       mm = mm_for_maps(priv->task);
+       if (!mm)
+               return NULL;
        down_read(&mm->mmap_sem);
 
+       tail_vma = get_gate_vma(priv->task);
+       priv->tail_vma = tail_vma;
+
        /* Start with last addr hint */
-       if (last_addr && (map = find_vma(mm, last_addr))) {
-               map = map->vm_next;
+       vma = find_vma(mm, last_addr);
+       if (last_addr && vma) {
+               vma = vma->vm_next;
                goto out;
        }
 
        /*
-        * Check the map index is within the range and do
+        * Check the vma index is within the range and do
         * sequential scan until m_index.
         */
-       map = NULL;
+       vma = NULL;
        if ((unsigned long)l < mm->map_count) {
-               map = mm->mmap;
-               while (l-- && map)
-                       map = map->vm_next;
+               vma = mm->mmap;
+               while (l-- && vma)
+                       vma = vma->vm_next;
                goto out;
        }
 
        if (l != mm->map_count)
-               tail_map = NULL; /* After gate map */
+               tail_vma = NULL; /* After gate vma */
 
 out:
-       if (map)
-               return map;
+       if (vma)
+               return vma;
 
-       /* End of maps has reached */
-       m->version = (tail_map != NULL)? 0: -1UL;
+       /* End of vmas has been reached */
+       m->version = (tail_vma != NULL)? 0: -1UL;
        up_read(&mm->mmap_sem);
        mmput(mm);
-       return tail_map;
+       return tail_vma;
+}
+
+static void *m_next(struct seq_file *m, void *v, loff_t *pos)
+{
+       struct proc_maps_private *priv = m->private;
+       struct vm_area_struct *vma = v;
+       struct vm_area_struct *tail_vma = priv->tail_vma;
+
+       (*pos)++;
+       if (vma && (vma != tail_vma) && vma->vm_next)
+               return vma->vm_next;
+       vma_stop(priv, vma);
+       return (vma != tail_vma)? tail_vma: NULL;
 }
 
 static void m_stop(struct seq_file *m, void *v)
 {
-       struct task_struct *task = m->private;
-       struct vm_area_struct *map = v;
-       if (map && map != get_gate_vma(task)) {
-               struct mm_struct *mm = map->vm_mm;
-               up_read(&mm->mmap_sem);
-               mmput(mm);
+       struct proc_maps_private *priv = m->private;
+       struct vm_area_struct *vma = v;
+
+       vma_stop(priv, vma);
+       if (priv->task)
+               put_task_struct(priv->task);
+}
+
+static int do_maps_open(struct inode *inode, struct file *file,
+                       const struct seq_operations *ops)
+{
+       struct proc_maps_private *priv;
+       int ret = -ENOMEM;
+       priv = kzalloc(sizeof(*priv), GFP_KERNEL);
+       if (priv) {
+               priv->pid = proc_pid(inode);
+               ret = seq_open(file, ops);
+               if (!ret) {
+                       struct seq_file *m = file->private_data;
+                       m->private = priv;
+               } else {
+                       kfree(priv);
+               }
        }
+       return ret;
 }
 
-static void *m_next(struct seq_file *m, void *v, loff_t *pos)
+static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 {
-       struct task_struct *task = m->private;
-       struct vm_area_struct *map = v;
-       struct vm_area_struct *tail_map = get_gate_vma(task);
+       struct mm_struct *mm = vma->vm_mm;
+       struct file *file = vma->vm_file;
+       int flags = vma->vm_flags;
+       unsigned long ino = 0;
+       unsigned long long pgoff = 0;
+       unsigned long start;
+       dev_t dev = 0;
+       int len;
 
-       (*pos)++;
-       if (map && (map != tail_map) && map->vm_next)
-               return map->vm_next;
-       m_stop(m, v);
-       return (map != tail_map)? tail_map: NULL;
+       if (file) {
+               struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
+               dev = inode->i_sb->s_dev;
+               ino = inode->i_ino;
+               pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
+       }
+
+       /* We don't show the stack guard page in /proc/maps */
+       start = vma->vm_start;
+       if (vma->vm_flags & VM_GROWSDOWN)
+               if (!vma_stack_continue(vma->vm_prev, vma->vm_start))
+                       start += PAGE_SIZE;
+
+       seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
+                       start,
+                       vma->vm_end,
+                       flags & VM_READ ? 'r' : '-',
+                       flags & VM_WRITE ? 'w' : '-',
+                       flags & VM_EXEC ? 'x' : '-',
+                       flags & VM_MAYSHARE ? 's' : 'p',
+                       pgoff,
+                       MAJOR(dev), MINOR(dev), ino, &len);
+
+       /*
+        * Print the dentry name for named mappings, and a
+        * special [heap] marker for the heap:
+        */
+       if (file) {
+               pad_len_spaces(m, len);
+               seq_path(m, &file->f_path, "\n");
+       } else {
+               const char *name = arch_vma_name(vma);
+               if (!name) {
+                       if (mm) {
+                               if (vma->vm_start <= mm->start_brk &&
+                                               vma->vm_end >= mm->brk) {
+                                       name = "[heap]";
+                               } else if (vma->vm_start <= mm->start_stack &&
+                                          vma->vm_end >= mm->start_stack) {
+                                       name = "[stack]";
+                               }
+                       } else {
+                               name = "[vdso]";
+                       }
+               }
+               if (name) {
+                       pad_len_spaces(m, len);
+                       seq_puts(m, name);
+               }
+       }
+       seq_putc(m, '\n');
+}
+
+static int show_map(struct seq_file *m, void *v)
+{
+       struct vm_area_struct *vma = v;
+       struct proc_maps_private *priv = m->private;
+       struct task_struct *task = priv->task;
+
+       show_map_vma(m, vma);
+
+       if (m->count < m->size)  /* vma is copied successfully */
+               m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
+       return 0;
 }
 
-struct seq_operations proc_pid_maps_op = {
+static const struct seq_operations proc_pid_maps_op = {
        .start  = m_start,
        .next   = m_next,
        .stop   = m_stop,
        .show   = show_map
 };
 
-#ifdef CONFIG_NUMA
+static int maps_open(struct inode *inode, struct file *file)
+{
+       return do_maps_open(inode, file, &proc_pid_maps_op);
+}
 
-struct numa_maps {
-       unsigned long pages;
-       unsigned long anon;
-       unsigned long mapped;
-       unsigned long mapcount_max;
-       unsigned long node[MAX_NUMNODES];
+const struct file_operations proc_maps_operations = {
+       .open           = maps_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = seq_release_private,
 };
 
 /*
- * Calculate numa node maps for a vma
+ * Proportional Set Size(PSS): my share of RSS.
+ *
+ * PSS of a process is the count of pages it has in memory, where each
+ * page is divided by the number of processes sharing it.  So if a
+ * process has 1000 pages all to itself, and 1000 shared with one other
+ * process, its PSS will be 1500.
+ *
+ * To keep (accumulated) division errors low, we adopt a 64bit
+ * fixed-point pss counter to minimize division errors. So (pss >>
+ * PSS_SHIFT) would be the real byte count.
+ *
+ * A shift of 12 before division means (assuming 4K page size):
+ *     - 1M 3-user-pages add up to 8KB errors;
+ *     - supports mapcount up to 2^24, or 16M;
+ *     - supports PSS up to 2^52 bytes, or 4PB.
  */
-static struct numa_maps *get_numa_maps(const struct vm_area_struct *vma)
+#define PSS_SHIFT 12
+
+#ifdef CONFIG_PROC_PAGE_MONITOR
+struct mem_size_stats {
+       struct vm_area_struct *vma;
+       unsigned long resident;
+       unsigned long shared_clean;
+       unsigned long shared_dirty;
+       unsigned long private_clean;
+       unsigned long private_dirty;
+       unsigned long referenced;
+       unsigned long swap;
+       u64 pss;
+};
+
+static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
+                          struct mm_walk *walk)
 {
+       struct mem_size_stats *mss = walk->private;
+       struct vm_area_struct *vma = mss->vma;
+       pte_t *pte, ptent;
+       spinlock_t *ptl;
        struct page *page;
-       unsigned long vaddr;
-       struct mm_struct *mm = vma->vm_mm;
-       int i;
-       struct numa_maps *md = kmalloc(sizeof(struct numa_maps), GFP_KERNEL);
+       int mapcount;
 
-       if (!md)
-               return NULL;
-       md->pages = 0;
-       md->anon = 0;
-       md->mapped = 0;
-       md->mapcount_max = 0;
-       for_each_node(i)
-               md->node[i] =0;
-
-       spin_lock(&mm->page_table_lock);
-       for (vaddr = vma->vm_start; vaddr < vma->vm_end; vaddr += PAGE_SIZE) {
-               page = follow_page(mm, vaddr, 0);
-               if (page) {
-                       int count = page_mapcount(page);
-
-                       if (count)
-                               md->mapped++;
-                       if (count > md->mapcount_max)
-                               md->mapcount_max = count;
-                       md->pages++;
-                       if (PageAnon(page))
-                               md->anon++;
-                       md->node[page_to_nid(page)]++;
+       pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
+       for (; addr != end; pte++, addr += PAGE_SIZE) {
+               ptent = *pte;
+
+               if (is_swap_pte(ptent)) {
+                       mss->swap += PAGE_SIZE;
+                       continue;
+               }
+
+               if (!pte_present(ptent))
+                       continue;
+
+               page = vm_normal_page(vma, addr, ptent);
+               if (!page)
+                       continue;
+
+               mss->resident += PAGE_SIZE;
+               /* Accumulate the size in pages that have been accessed. */
+               if (pte_young(ptent) || PageReferenced(page))
+                       mss->referenced += PAGE_SIZE;
+               mapcount = page_mapcount(page);
+               if (mapcount >= 2) {
+                       if (pte_dirty(ptent) || PageDirty(page))
+                               mss->shared_dirty += PAGE_SIZE;
+                       else
+                               mss->shared_clean += PAGE_SIZE;
+                       mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
+               } else {
+                       if (pte_dirty(ptent) || PageDirty(page))
+                               mss->private_dirty += PAGE_SIZE;
+                       else
+                               mss->private_clean += PAGE_SIZE;
+                       mss->pss += (PAGE_SIZE << PSS_SHIFT);
                }
        }
-       spin_unlock(&mm->page_table_lock);
-       return md;
+       pte_unmap_unlock(pte - 1, ptl);
+       cond_resched();
+       return 0;
 }
 
-static int show_numa_map(struct seq_file *m, void *v)
+static int show_smap(struct seq_file *m, void *v)
 {
-       struct task_struct *task = m->private;
+       struct proc_maps_private *priv = m->private;
+       struct task_struct *task = priv->task;
        struct vm_area_struct *vma = v;
-       struct mempolicy *pol;
-       struct numa_maps *md;
-       struct zone **z;
-       int n;
-       int first;
-
-       if (!vma->vm_mm)
-               return 0;
-
-       md = get_numa_maps(vma);
-       if (!md)
-               return 0;
-
-       seq_printf(m, "%08lx", vma->vm_start);
-       pol = get_vma_policy(task, vma, vma->vm_start);
-       /* Print policy */
-       switch (pol->policy) {
-       case MPOL_PREFERRED:
-               seq_printf(m, " prefer=%d", pol->v.preferred_node);
-               break;
-       case MPOL_BIND:
-               seq_printf(m, " bind={");
-               first = 1;
-               for (z = pol->v.zonelist->zones; *z; z++) {
-
-                       if (!first)
-                               seq_putc(m, ',');
-                       else
-                               first = 0;
-                       seq_printf(m, "%d/%s", (*z)->zone_pgdat->node_id,
-                                       (*z)->name);
-               }
-               seq_putc(m, '}');
-               break;
-       case MPOL_INTERLEAVE:
-               seq_printf(m, " interleave={");
-               first = 1;
-               for_each_node(n) {
-                       if (test_bit(n, pol->v.nodes)) {
-                               if (!first)
-                                       seq_putc(m,',');
-                               else
-                                       first = 0;
-                               seq_printf(m, "%d",n);
-                       }
-               }
-               seq_putc(m, '}');
-               break;
-       default:
-               seq_printf(m," default");
-               break;
-       }
-       seq_printf(m, " MaxRef=%lu Pages=%lu Mapped=%lu",
-                       md->mapcount_max, md->pages, md->mapped);
-       if (md->anon)
-               seq_printf(m," Anon=%lu",md->anon);
-
-       for_each_online_node(n) {
-               if (md->node[n])
-                       seq_printf(m, " N%d=%lu", n, md->node[n]);
-       }
-       seq_putc(m, '\n');
-       kfree(md);
+       struct mem_size_stats mss;
+       struct mm_walk smaps_walk = {
+               .pmd_entry = smaps_pte_range,
+               .mm = vma->vm_mm,
+               .private = &mss,
+       };
+
+       memset(&mss, 0, sizeof mss);
+       mss.vma = vma;
+       /* mmap_sem is held in m_start */
+       if (vma->vm_mm && !is_vm_hugetlb_page(vma))
+               walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
+
+       show_map_vma(m, vma);
+
+       seq_printf(m,
+                  "Size:           %8lu kB\n"
+                  "Rss:            %8lu kB\n"
+                  "Pss:            %8lu kB\n"
+                  "Shared_Clean:   %8lu kB\n"
+                  "Shared_Dirty:   %8lu kB\n"
+                  "Private_Clean:  %8lu kB\n"
+                  "Private_Dirty:  %8lu kB\n"
+                  "Referenced:     %8lu kB\n"
+                  "Swap:           %8lu kB\n"
+                  "KernelPageSize: %8lu kB\n"
+                  "MMUPageSize:    %8lu kB\n",
+                  (vma->vm_end - vma->vm_start) >> 10,
+                  mss.resident >> 10,
+                  (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
+                  mss.shared_clean  >> 10,
+                  mss.shared_dirty  >> 10,
+                  mss.private_clean >> 10,
+                  mss.private_dirty >> 10,
+                  mss.referenced >> 10,
+                  mss.swap >> 10,
+                  vma_kernel_pagesize(vma) >> 10,
+                  vma_mmu_pagesize(vma) >> 10);
+
        if (m->count < m->size)  /* vma is copied successfully */
                m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
        return 0;
 }
 
-struct seq_operations proc_pid_numa_maps_op = {
+static const struct seq_operations proc_pid_smaps_op = {
        .start  = m_start,
        .next   = m_next,
        .stop   = m_stop,
-       .show   = show_numa_map
+       .show   = show_smap
+};
+
+static int smaps_open(struct inode *inode, struct file *file)
+{
+       return do_maps_open(inode, file, &proc_pid_smaps_op);
+}
+
+const struct file_operations proc_smaps_operations = {
+       .open           = smaps_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = seq_release_private,
+};
+
+static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
+                               unsigned long end, struct mm_walk *walk)
+{
+       struct vm_area_struct *vma = walk->private;
+       pte_t *pte, ptent;
+       spinlock_t *ptl;
+       struct page *page;
+
+       pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
+       for (; addr != end; pte++, addr += PAGE_SIZE) {
+               ptent = *pte;
+               if (!pte_present(ptent))
+                       continue;
+
+               page = vm_normal_page(vma, addr, ptent);
+               if (!page)
+                       continue;
+
+               /* Clear accessed and referenced bits. */
+               ptep_test_and_clear_young(vma, addr, pte);
+               ClearPageReferenced(page);
+       }
+       pte_unmap_unlock(pte - 1, ptl);
+       cond_resched();
+       return 0;
+}
+
+#define CLEAR_REFS_ALL 1
+#define CLEAR_REFS_ANON 2
+#define CLEAR_REFS_MAPPED 3
+
+static ssize_t clear_refs_write(struct file *file, const char __user *buf,
+                               size_t count, loff_t *ppos)
+{
+       struct task_struct *task;
+       char buffer[PROC_NUMBUF];
+       struct mm_struct *mm;
+       struct vm_area_struct *vma;
+       long type;
+
+       memset(buffer, 0, sizeof(buffer));
+       if (count > sizeof(buffer) - 1)
+               count = sizeof(buffer) - 1;
+       if (copy_from_user(buffer, buf, count))
+               return -EFAULT;
+       if (strict_strtol(strstrip(buffer), 10, &type))
+               return -EINVAL;
+       if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
+               return -EINVAL;
+       task = get_proc_task(file->f_path.dentry->d_inode);
+       if (!task)
+               return -ESRCH;
+       mm = get_task_mm(task);
+       if (mm) {
+               struct mm_walk clear_refs_walk = {
+                       .pmd_entry = clear_refs_pte_range,
+                       .mm = mm,
+               };
+               down_read(&mm->mmap_sem);
+               for (vma = mm->mmap; vma; vma = vma->vm_next) {
+                       clear_refs_walk.private = vma;
+                       if (is_vm_hugetlb_page(vma))
+                               continue;
+                       /*
+                        * Writing 1 to /proc/pid/clear_refs affects all pages.
+                        *
+                        * Writing 2 to /proc/pid/clear_refs only affects
+                        * Anonymous pages.
+                        *
+                        * Writing 3 to /proc/pid/clear_refs only affects file
+                        * mapped pages.
+                        */
+                       if (type == CLEAR_REFS_ANON && vma->vm_file)
+                               continue;
+                       if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
+                               continue;
+                       walk_page_range(vma->vm_start, vma->vm_end,
+                                       &clear_refs_walk);
+               }
+               flush_tlb_mm(mm);
+               up_read(&mm->mmap_sem);
+               mmput(mm);
+       }
+       put_task_struct(task);
+
+       return count;
+}
+
+const struct file_operations proc_clear_refs_operations = {
+       .write          = clear_refs_write,
+       .llseek         = noop_llseek,
+};
+
+struct pagemapread {
+       int pos, len;
+       u64 *buffer;
+};
+
+#define PM_ENTRY_BYTES      sizeof(u64)
+#define PM_STATUS_BITS      3
+#define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
+#define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
+#define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
+#define PM_PSHIFT_BITS      6
+#define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
+#define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
+#define PM_PSHIFT(x)        (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
+#define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
+#define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
+
+#define PM_PRESENT          PM_STATUS(4LL)
+#define PM_SWAP             PM_STATUS(2LL)
+#define PM_NOT_PRESENT      PM_PSHIFT(PAGE_SHIFT)
+#define PM_END_OF_BUFFER    1
+
+static int add_to_pagemap(unsigned long addr, u64 pfn,
+                         struct pagemapread *pm)
+{
+       pm->buffer[pm->pos++] = pfn;
+       if (pm->pos >= pm->len)
+               return PM_END_OF_BUFFER;
+       return 0;
+}
+
+static int pagemap_pte_hole(unsigned long start, unsigned long end,
+                               struct mm_walk *walk)
+{
+       struct pagemapread *pm = walk->private;
+       unsigned long addr;
+       int err = 0;
+       for (addr = start; addr < end; addr += PAGE_SIZE) {
+               err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
+               if (err)
+                       break;
+       }
+       return err;
+}
+
+static u64 swap_pte_to_pagemap_entry(pte_t pte)
+{
+       swp_entry_t e = pte_to_swp_entry(pte);
+       return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
+}
+
+static u64 pte_to_pagemap_entry(pte_t pte)
+{
+       u64 pme = 0;
+       if (is_swap_pte(pte))
+               pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
+                       | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
+       else if (pte_present(pte))
+               pme = PM_PFRAME(pte_pfn(pte))
+                       | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
+       return pme;
+}
+
+static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
+                            struct mm_walk *walk)
+{
+       struct vm_area_struct *vma;
+       struct pagemapread *pm = walk->private;
+       pte_t *pte;
+       int err = 0;
+
+       /* find the first VMA at or above 'addr' */
+       vma = find_vma(walk->mm, addr);
+       for (; addr != end; addr += PAGE_SIZE) {
+               u64 pfn = PM_NOT_PRESENT;
+
+               /* check to see if we've left 'vma' behind
+                * and need a new, higher one */
+               if (vma && (addr >= vma->vm_end))
+                       vma = find_vma(walk->mm, addr);
+
+               /* check that 'vma' actually covers this address,
+                * and that it isn't a huge page vma */
+               if (vma && (vma->vm_start <= addr) &&
+                   !is_vm_hugetlb_page(vma)) {
+                       pte = pte_offset_map(pmd, addr);
+                       pfn = pte_to_pagemap_entry(*pte);
+                       /* unmap before userspace copy */
+                       pte_unmap(pte);
+               }
+               err = add_to_pagemap(addr, pfn, pm);
+               if (err)
+                       return err;
+       }
+
+       cond_resched();
+
+       return err;
+}
+
+#ifdef CONFIG_HUGETLB_PAGE
+static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
+{
+       u64 pme = 0;
+       if (pte_present(pte))
+               pme = PM_PFRAME(pte_pfn(pte) + offset)
+                       | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
+       return pme;
+}
+
+/* This function walks within one hugetlb entry in the single call */
+static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
+                                unsigned long addr, unsigned long end,
+                                struct mm_walk *walk)
+{
+       struct pagemapread *pm = walk->private;
+       int err = 0;
+       u64 pfn;
+
+       for (; addr != end; addr += PAGE_SIZE) {
+               int offset = (addr & ~hmask) >> PAGE_SHIFT;
+               pfn = huge_pte_to_pagemap_entry(*pte, offset);
+               err = add_to_pagemap(addr, pfn, pm);
+               if (err)
+                       return err;
+       }
+
+       cond_resched();
+
+       return err;
+}
+#endif /* HUGETLB_PAGE */
+
+/*
+ * /proc/pid/pagemap - an array mapping virtual pages to pfns
+ *
+ * For each page in the address space, this file contains one 64-bit entry
+ * consisting of the following:
+ *
+ * Bits 0-55  page frame number (PFN) if present
+ * Bits 0-4   swap type if swapped
+ * Bits 5-55  swap offset if swapped
+ * Bits 55-60 page shift (page size = 1<<page shift)
+ * Bit  61    reserved for future use
+ * Bit  62    page swapped
+ * Bit  63    page present
+ *
+ * If the page is not present but in swap, then the PFN contains an
+ * encoding of the swap file number and the page's offset into the
+ * swap. Unmapped pages return a null PFN. This allows determining
+ * precisely which pages are mapped (or in swap) and comparing mapped
+ * pages between processes.
+ *
+ * Efficient users of this interface will use /proc/pid/maps to
+ * determine which areas of memory are actually mapped and llseek to
+ * skip over unmapped regions.
+ */
+#define PAGEMAP_WALK_SIZE      (PMD_SIZE)
+static ssize_t pagemap_read(struct file *file, char __user *buf,
+                           size_t count, loff_t *ppos)
+{
+       struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
+       struct mm_struct *mm;
+       struct pagemapread pm;
+       int ret = -ESRCH;
+       struct mm_walk pagemap_walk = {};
+       unsigned long src;
+       unsigned long svpfn;
+       unsigned long start_vaddr;
+       unsigned long end_vaddr;
+       int copied = 0;
+
+       if (!task)
+               goto out;
+
+       ret = -EACCES;
+       if (!ptrace_may_access(task, PTRACE_MODE_READ))
+               goto out_task;
+
+       ret = -EINVAL;
+       /* file position must be aligned */
+       if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
+               goto out_task;
+
+       ret = 0;
+
+       if (!count)
+               goto out_task;
+
+       mm = get_task_mm(task);
+       if (!mm)
+               goto out_task;
+
+       pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
+       pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
+       ret = -ENOMEM;
+       if (!pm.buffer)
+               goto out_mm;
+
+       pagemap_walk.pmd_entry = pagemap_pte_range;
+       pagemap_walk.pte_hole = pagemap_pte_hole;
+#ifdef CONFIG_HUGETLB_PAGE
+       pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
+#endif
+       pagemap_walk.mm = mm;
+       pagemap_walk.private = &pm;
+
+       src = *ppos;
+       svpfn = src / PM_ENTRY_BYTES;
+       start_vaddr = svpfn << PAGE_SHIFT;
+       end_vaddr = TASK_SIZE_OF(task);
+
+       /* watch out for wraparound */
+       if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
+               start_vaddr = end_vaddr;
+
+       /*
+        * The odds are that this will stop walking way
+        * before end_vaddr, because the length of the
+        * user buffer is tracked in "pm", and the walk
+        * will stop when we hit the end of the buffer.
+        */
+       ret = 0;
+       while (count && (start_vaddr < end_vaddr)) {
+               int len;
+               unsigned long end;
+
+               pm.pos = 0;
+               end = start_vaddr + PAGEMAP_WALK_SIZE;
+               /* overflow ? */
+               if (end < start_vaddr || end > end_vaddr)
+                       end = end_vaddr;
+               down_read(&mm->mmap_sem);
+               ret = walk_page_range(start_vaddr, end, &pagemap_walk);
+               up_read(&mm->mmap_sem);
+               start_vaddr = end;
+
+               len = min(count, PM_ENTRY_BYTES * pm.pos);
+               if (copy_to_user(buf, pm.buffer, len)) {
+                       ret = -EFAULT;
+                       goto out_free;
+               }
+               copied += len;
+               buf += len;
+               count -= len;
+       }
+       *ppos += copied;
+       if (!ret || ret == PM_END_OF_BUFFER)
+               ret = copied;
+
+out_free:
+       kfree(pm.buffer);
+out_mm:
+       mmput(mm);
+out_task:
+       put_task_struct(task);
+out:
+       return ret;
+}
+
+const struct file_operations proc_pagemap_operations = {
+       .llseek         = mem_lseek, /* borrow this */
+       .read           = pagemap_read,
+};
+#endif /* CONFIG_PROC_PAGE_MONITOR */
+
+#ifdef CONFIG_NUMA
+extern int show_numa_map(struct seq_file *m, void *v);
+
+static const struct seq_operations proc_pid_numa_maps_op = {
+        .start  = m_start,
+        .next   = m_next,
+        .stop   = m_stop,
+        .show   = show_numa_map,
+};
+
+static int numa_maps_open(struct inode *inode, struct file *file)
+{
+       return do_maps_open(inode, file, &proc_pid_numa_maps_op);
+}
+
+const struct file_operations proc_numa_maps_operations = {
+       .open           = numa_maps_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = seq_release_private,
 };
 #endif