ext4-page-mkwrite.patch

   1 ext4: Use page_mkwrite vma_operations to get mmap write notification.
   2
   3 From: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
   4
   5 We would like to get notified when we are doing a write on mmap section.
   6 This is needed with respect to preallocated area. We split the preallocated
   7 area into initialzed extent and uninitialzed extent in the call back. This
   8 let us handle ENOSPC better. Otherwise we get ENOSPC in the writepage and
   9 that would result in data loss. The changes are also needed to handle ENOSPC
  10 when writing to an mmap section of files with holes.
  11
  12 Acked-by: Jan Kara <jack@suse.cz>
  13 Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
  14 Signed-off-by: Mingming Cao <cmm@us.ibm.com>
  15 Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
  16 ---
  17
  18  fs/ext4/ext4.h  |    1
  19  fs/ext4/file.c  |   19 ++++++++++++++++-
  20  fs/ext4/inode.c |   61 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
  21  3 files changed, 80 insertions(+), 1 deletion(-)
  22
  23
  24 Index: linux-2.6.26-rc6/fs/ext4/ext4.h
  25 ===================================================================
  26 --- linux-2.6.26-rc6.orig/fs/ext4/ext4.h        2008-06-17 10:43:32.000000000 -0700
  27 +++ linux-2.6.26-rc6/fs/ext4/ext4.h     2008-06-17 10:43:34.000000000 -0700
  28 @@ -1063,6 +1063,7 @@ extern void ext4_set_aops(struct inode *
  29  extern int ext4_writepage_trans_blocks(struct inode *);
  30  extern int ext4_block_truncate_page(handle_t *handle, struct page *page,
  31                 struct address_space *mapping, loff_t from);
  32 +extern int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page);
  33
  34  /* ioctl.c */
  35  extern long ext4_ioctl(struct file *, unsigned int, unsigned long);
  36 Index: linux-2.6.26-rc6/fs/ext4/file.c
  37 ===================================================================
  38 --- linux-2.6.26-rc6.orig/fs/ext4/file.c        2008-06-17 10:43:18.000000000 -0700
  39 +++ linux-2.6.26-rc6/fs/ext4/file.c     2008-06-17 10:43:34.000000000 -0700
  40 @@ -123,6 +123,23 @@ force_commit:
  41         return ret;
  42  }
  43
  44 +static struct vm_operations_struct ext4_file_vm_ops = {
  45 +       .fault          = filemap_fault,
  46 +       .page_mkwrite   = ext4_page_mkwrite,
  47 +};
  48 +
  49 +static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
  50 +{
  51 +       struct address_space *mapping = file->f_mapping;
  52 +
  53 +       if (!mapping->a_ops->readpage)
  54 +               return -ENOEXEC;
  55 +       file_accessed(file);
  56 +       vma->vm_ops = &ext4_file_vm_ops;
  57 +       vma->vm_flags |= VM_CAN_NONLINEAR;
  58 +       return 0;
  59 +}
  60 +
  61  const struct file_operations ext4_file_operations = {
  62         .llseek         = generic_file_llseek,
  63         .read           = do_sync_read,
  64 @@ -133,7 +150,7 @@ const struct file_operations ext4_file_o
  65  #ifdef CONFIG_COMPAT
  66         .compat_ioctl   = ext4_compat_ioctl,
  67  #endif
  68 -       .mmap           = generic_file_mmap,
  69 +       .mmap           = ext4_file_mmap,
  70         .open           = generic_file_open,
  71         .release        = ext4_release_file,
  72         .fsync          = ext4_sync_file,
  73 Index: linux-2.6.26-rc6/fs/ext4/inode.c
  74 ===================================================================
  75 --- linux-2.6.26-rc6.orig/fs/ext4/inode.c       2008-06-17 10:43:31.000000000 -0700
  76 +++ linux-2.6.26-rc6/fs/ext4/inode.c    2008-06-17 10:43:34.000000000 -0700
  77 @@ -3542,3 +3542,64 @@ int ext4_change_inode_journal_flag(struc
  78
  79         return err;
  80  }
  81 +
  82 +static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
  83 +{
  84 +       return !buffer_mapped(bh);
  85 +}
  86 +
  87 +int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
  88 +{
  89 +       loff_t size;
  90 +       unsigned long len;
  91 +       int ret = -EINVAL;
  92 +       struct file *file = vma->vm_file;
  93 +       struct inode *inode = file->f_path.dentry->d_inode;
  94 +       struct address_space *mapping = inode->i_mapping;
  95 +
  96 +       /*
  97 +        * Get i_alloc_sem to stop truncates messing with the inode. We cannot
  98 +        * get i_mutex because we are already holding mmap_sem.
  99 +        */
 100 +       down_read(&inode->i_alloc_sem);
 101 +       size = i_size_read(inode);
 102 +       if (page->mapping != mapping || size <= page_offset(page)
 103 +           || !PageUptodate(page)) {
 104 +               /* page got truncated from under us? */
 105 +               goto out_unlock;
 106 +       }
 107 +       ret = 0;
 108 +       if (PageMappedToDisk(page))
 109 +               goto out_unlock;
 110 +
 111 +       if (page->index == size >> PAGE_CACHE_SHIFT)
 112 +               len = size & ~PAGE_CACHE_MASK;
 113 +       else
 114 +               len = PAGE_CACHE_SIZE;
 115 +
 116 +       if (page_has_buffers(page)) {
 117 +               /* return if we have all the buffers mapped */
 118 +               if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
 119 +                                      ext4_bh_unmapped))
 120 +                       goto out_unlock;
 121 +       }
 122 +       /*
 123 +        * OK, we need to fill the hole... Do write_begin write_end
 124 +        * to do block allocation/reservation.We are not holding
 125 +        * inode.i__mutex here. That allow * parallel write_begin,
 126 +        * write_end call. lock_page prevent this from happening
 127 +        * on the same page though
 128 +        */
 129 +       ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
 130 +                       len, AOP_FLAG_UNINTERRUPTIBLE, &page, NULL);
 131 +       if (ret < 0)
 132 +               goto out_unlock;
 133 +       ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
 134 +                       len, len, page, NULL);
 135 +       if (ret < 0)
 136 +               goto out_unlock;
 137 +       ret = 0;
 138 +out_unlock:
 139 +       up_read(&inode->i_alloc_sem);
 140 +       return ret;
 141 +}