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 bbd9b14..871e25e 100644 (file)
@@ -4,8 +4,8 @@
 #include <linux/seq_file.h>
 #include <linux/highmem.h>
 #include <linux/ptrace.h>
+#include <linux/slab.h>
 #include <linux/pagemap.h>
-#include <linux/ptrace.h>
 #include <linux/mempolicy.h>
 #include <linux/swap.h>
 #include <linux/swapops.h>
@@ -15,9 +15,9 @@
 #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;
 
        /*
@@ -37,7 +37,8 @@ char *task_mem(struct mm_struct *mm, char *buffer)
        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"
@@ -47,7 +48,8 @@ char *task_mem(struct mm_struct *mm, char *buffer)
                "VmStk:\t%8lu kB\n"
                "VmExe:\t%8lu kB\n"
                "VmLib:\t%8lu kB\n"
-               "VmPTE:\t%8lu kB\n",
+               "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),
@@ -55,8 +57,8 @@ char *task_mem(struct mm_struct *mm, char *buffer)
                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)
@@ -67,48 +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)
 {
-       *shared = get_mm_counter(mm, file_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 = *shared + get_mm_counter(mm, anon_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 = get_proc_task(inode);
-       struct mm_struct * mm = NULL;
-
-       if (task) {
-               mm = get_task_mm(task);
-               put_task_struct(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_path.mnt);
-               *dentry = dget(vma->vm_file->f_path.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;
@@ -155,6 +123,7 @@ static void *m_start(struct seq_file *m, loff_t *pos)
        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;
@@ -216,7 +185,7 @@ static void m_stop(struct seq_file *m, void *v)
 }
 
 static int do_maps_open(struct inode *inode, struct file *file,
-                       struct seq_operations *ops)
+                       const struct seq_operations *ops)
 {
        struct proc_maps_private *priv;
        int ret = -ENOMEM;
@@ -234,35 +203,38 @@ static int do_maps_open(struct inode *inode, struct file *file,
        return ret;
 }
 
-static int show_map(struct seq_file *m, void *v)
+static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 {
-       struct proc_maps_private *priv = m->private;
-       struct task_struct *task = priv->task;
-       struct vm_area_struct *vma = v;
        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;
 
-       if (maps_protect && !ptrace_may_attach(task))
-               return -EACCES;
-
        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;
        }
 
-       seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
-                       vma->vm_start,
+       /* 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',
-                       vma->vm_pgoff << PAGE_SHIFT,
+                       pgoff,
                        MAJOR(dev), MINOR(dev), ino, &len);
 
        /*
@@ -271,7 +243,7 @@ static int show_map(struct seq_file *m, void *v)
         */
        if (file) {
                pad_len_spaces(m, len);
-               seq_path(m, file->f_path.mnt, file->f_path.dentry, "\n");
+               seq_path(m, &file->f_path, "\n");
        } else {
                const char *name = arch_vma_name(vma);
                if (!name) {
@@ -293,13 +265,22 @@ static int show_map(struct seq_file *m, void *v)
                }
        }
        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;
 }
 
-static 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,
@@ -337,8 +318,8 @@ const struct file_operations proc_maps_operations = {
  */
 #define PSS_SHIFT 12
 
-struct mem_size_stats
-{
+#ifdef CONFIG_PROC_PAGE_MONITOR
+struct mem_size_stats {
        struct vm_area_struct *vma;
        unsigned long resident;
        unsigned long shared_clean;
@@ -346,13 +327,14 @@ struct mem_size_stats
        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,
-                          void *private)
+                          struct mm_walk *walk)
 {
-       struct mem_size_stats *mss = private;
+       struct mem_size_stats *mss = walk->private;
        struct vm_area_struct *vma = mss->vma;
        pte_t *pte, ptent;
        spinlock_t *ptl;
@@ -362,27 +344,32 @@ static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
        for (; addr != end; pte++, addr += PAGE_SIZE) {
                ptent = *pte;
-               if (!pte_present(ptent))
+
+               if (is_swap_pte(ptent)) {
+                       mss->swap += PAGE_SIZE;
                        continue;
+               }
 
-               mss->resident += PAGE_SIZE;
+               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))
+                       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))
+                       if (pte_dirty(ptent) || PageDirty(page))
                                mss->private_dirty += PAGE_SIZE;
                        else
                                mss->private_clean += PAGE_SIZE;
@@ -394,23 +381,25 @@ static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
        return 0;
 }
 
-static struct mm_walk smaps_walk = { .pmd_entry = smaps_pte_range };
-
 static int show_smap(struct seq_file *m, void *v)
 {
+       struct proc_maps_private *priv = m->private;
+       struct task_struct *task = priv->task;
        struct vm_area_struct *vma = v;
        struct mem_size_stats mss;
-       int ret;
+       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_mm, vma->vm_start, vma->vm_end,
-                               &smaps_walk, &mss);
+               walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
 
-       ret = show_map(m, v);
-       if (ret)
-               return ret;
+       show_map_vma(m, vma);
 
        seq_printf(m,
                   "Size:           %8lu kB\n"
@@ -420,7 +409,10 @@ static int show_smap(struct seq_file *m, void *v)
                   "Shared_Dirty:   %8lu kB\n"
                   "Private_Clean:  %8lu kB\n"
                   "Private_Dirty:  %8lu kB\n"
-                  "Referenced:     %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)),
@@ -428,12 +420,17 @@ static int show_smap(struct seq_file *m, void *v)
                   mss.shared_dirty  >> 10,
                   mss.private_clean >> 10,
                   mss.private_dirty >> 10,
-                  mss.referenced >> 10);
+                  mss.referenced >> 10,
+                  mss.swap >> 10,
+                  vma_kernel_pagesize(vma) >> 10,
+                  vma_mmu_pagesize(vma) >> 10);
 
-       return ret;
+       if (m->count < m->size)  /* vma is copied successfully */
+               m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
+       return 0;
 }
 
-static struct seq_operations proc_pid_smaps_op = {
+static const struct seq_operations proc_pid_smaps_op = {
        .start  = m_start,
        .next   = m_next,
        .stop   = m_stop,
@@ -453,9 +450,9 @@ const struct file_operations proc_smaps_operations = {
 };
 
 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
-                               unsigned long end, void *private)
+                               unsigned long end, struct mm_walk *walk)
 {
-       struct vm_area_struct *vma = private;
+       struct vm_area_struct *vma = walk->private;
        pte_t *pte, ptent;
        spinlock_t *ptl;
        struct page *page;
@@ -479,87 +476,107 @@ static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
        return 0;
 }
 
-static struct mm_walk clear_refs_walk = { .pmd_entry = clear_refs_pte_range };
+#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], *end;
+       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 (!simple_strtol(buffer, &end, 0))
+       if (strict_strtol(strstrip(buffer), 10, &type))
+               return -EINVAL;
+       if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
                return -EINVAL;
-       if (*end == '\n')
-               end++;
        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)
-                       if (!is_vm_hugetlb_page(vma))
-                               walk_page_range(mm, vma->vm_start, vma->vm_end,
-                                               &clear_refs_walk, vma);
+               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);
-       if (end - buffer == 0)
-               return -EIO;
-       return end - buffer;
+
+       return count;
 }
 
 const struct file_operations proc_clear_refs_operations = {
        .write          = clear_refs_write,
+       .llseek         = noop_llseek,
 };
 
 struct pagemapread {
-       char __user *out, *end;
+       int pos, len;
+       u64 *buffer;
 };
 
-#define PM_ENTRY_BYTES sizeof(u64)
-#define PM_RESERVED_BITS    3
-#define PM_RESERVED_OFFSET  (64 - PM_RESERVED_BITS)
-#define PM_RESERVED_MASK    (((1LL<<PM_RESERVED_BITS)-1) << PM_RESERVED_OFFSET)
-#define PM_SPECIAL(nr)      (((nr) << PM_RESERVED_OFFSET) | PM_RESERVED_MASK)
-#define PM_NOT_PRESENT      PM_SPECIAL(1LL)
-#define PM_SWAP             PM_SPECIAL(2LL)
+#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)
 {
-       /*
-        * Make sure there's room in the buffer for an
-        * entire entry.  Otherwise, only copy part of
-        * the pfn.
-        */
-       if (pm->out + PM_ENTRY_BYTES >= pm->end) {
-               if (copy_to_user(pm->out, &pfn, pm->end - pm->out))
-                       return -EFAULT;
-               pm->out = pm->end;
+       pm->buffer[pm->pos++] = pfn;
+       if (pm->pos >= pm->len)
                return PM_END_OF_BUFFER;
-       }
-
-       if (put_user(pfn, pm->out))
-               return -EFAULT;
-       pm->out += PM_ENTRY_BYTES;
        return 0;
 }
 
 static int pagemap_pte_hole(unsigned long start, unsigned long end,
-                               void *private)
+                               struct mm_walk *walk)
 {
-       struct pagemapread *pm = private;
+       struct pagemapread *pm = walk->private;
        unsigned long addr;
        int err = 0;
        for (addr = start; addr < end; addr += PAGE_SIZE) {
@@ -570,28 +587,51 @@ static int pagemap_pte_hole(unsigned long start, unsigned long end,
        return err;
 }
 
-u64 swap_pte_to_pagemap_entry(pte_t pte)
+static u64 swap_pte_to_pagemap_entry(pte_t pte)
 {
        swp_entry_t e = pte_to_swp_entry(pte);
-       return PM_SWAP | swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
+       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,
-                            void *private)
+                            struct mm_walk *walk)
 {
-       struct pagemapread *pm = private;
+       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;
-               pte = pte_offset_map(pmd, addr);
-               if (is_swap_pte(*pte))
-                       pfn = swap_pte_to_pagemap_entry(*pte);
-               else if (pte_present(*pte))
-                       pfn = pte_pfn(*pte);
-               /* unmap so we're not in atomic when we copy to userspace */
-               pte_unmap(pte);
+
+               /* 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;
@@ -602,20 +642,56 @@ static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
        return err;
 }
 
-static struct mm_walk pagemap_walk = {
-       .pmd_entry = pagemap_pte_range,
-       .pte_hole = pagemap_pte_hole
-};
+#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 representing the corresponding physical page frame number
- * (PFN) if the page is present. If there is a swap entry for the
- * physical page, then an encoding of the swap file number and the
- * page's offset into the swap file are returned. If no page is
- * present at all, PM_NOT_PRESENT is returned. This allows determining
+ * 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.
  *
@@ -623,90 +699,103 @@ static struct mm_walk pagemap_walk = {
  * 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 page **pages, *page;
-       unsigned long uaddr, uend;
        struct mm_struct *mm;
        struct pagemapread pm;
-       int pagecount;
        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_attach(task))
-               goto out;
+       if (!ptrace_may_access(task, PTRACE_MODE_READ))
+               goto out_task;
 
        ret = -EINVAL;
        /* file position must be aligned */
-       if (*ppos % PM_ENTRY_BYTES)
-               goto out;
+       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;
+               goto out_task;
 
+       pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
+       pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
        ret = -ENOMEM;
-       uaddr = (unsigned long)buf & PAGE_MASK;
-       uend = (unsigned long)(buf + count);
-       pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE;
-       pages = kmalloc(pagecount * sizeof(struct page *), GFP_KERNEL);
-       if (!pages)
-               goto out_task;
+       if (!pm.buffer)
+               goto out_mm;
 
-       down_read(&current->mm->mmap_sem);
-       ret = get_user_pages(current, current->mm, uaddr, pagecount,
-                            1, 0, pages, NULL);
-       up_read(&current->mm->mmap_sem);
+       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;
 
-       if (ret < 0)
-               goto out_free;
+       src = *ppos;
+       svpfn = src / PM_ENTRY_BYTES;
+       start_vaddr = svpfn << PAGE_SHIFT;
+       end_vaddr = TASK_SIZE_OF(task);
 
-       pm.out = buf;
-       pm.end = buf + count;
+       /* watch out for wraparound */
+       if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
+               start_vaddr = end_vaddr;
 
-       if (!ptrace_may_attach(task)) {
-               ret = -EIO;
-       } else {
-               unsigned long src = *ppos;
-               unsigned long svpfn = src / PM_ENTRY_BYTES;
-               unsigned long start_vaddr = svpfn << PAGE_SHIFT;
-               unsigned long 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 = walk_page_range(mm, start_vaddr, end_vaddr,
-                                       &pagemap_walk, &pm);
-               if (ret == PM_END_OF_BUFFER)
-                       ret = 0;
-               /* don't need mmap_sem for these, but this looks cleaner */
-               *ppos += pm.out - buf;
-               if (!ret)
-                       ret = pm.out - buf;
-       }
+       /*
+        * 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;
 
-       for (; pagecount; pagecount--) {
-               page = pages[pagecount-1];
-               if (!PageReserved(page))
-                       SetPageDirty(page);
-               page_cache_release(page);
+               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;
        }
-       mmput(mm);
+       *ppos += copied;
+       if (!ret || ret == PM_END_OF_BUFFER)
+               ret = copied;
+
 out_free:
-       kfree(pages);
+       kfree(pm.buffer);
+out_mm:
+       mmput(mm);
 out_task:
        put_task_struct(task);
 out:
@@ -717,26 +806,16 @@ 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 int show_numa_map_checked(struct seq_file *m, void *v)
-{
-       struct proc_maps_private *priv = m->private;
-       struct task_struct *task = priv->task;
-
-       if (maps_protect && !ptrace_may_attach(task))
-               return -EACCES;
-
-       return show_numa_map(m, v);
-}
-
-static struct seq_operations proc_pid_numa_maps_op = {
+static const struct seq_operations proc_pid_numa_maps_op = {
         .start  = m_start,
         .next   = m_next,
         .stop   = m_stop,
-        .show   = show_numa_map_checked
+        .show   = show_numa_map,
 };
 
 static int numa_maps_open(struct inode *inode, struct file *file)