lkml.org 
[lkml]   [2008]   [Jun]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: Performance of ext4
    On Tue, 24 Jun 2008, Aneesh Kumar K.V wrote:

    > On Tue, Jun 24, 2008 at 08:37:21AM +0530, Aneesh Kumar K.V wrote:
    >> On Mon, Jun 23, 2008 at 05:31:32PM -0700, Mingming wrote:
    >>>
    >>> On Mon, 2008-06-23 at 23:15 +0530, Aneesh Kumar K.V wrote:
    >>>
    >>>> I found one place where we fail to update i_disksize. Can you try this
    >>>> patch ?
    >>>>
    >>>> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
    >>>> index 33f940b..9fa737f 100644
    >>>> --- a/fs/ext4/inode.c
    >>>> +++ b/fs/ext4/inode.c
    >>>> @@ -1620,7 +1620,10 @@ static int ext4_da_writepage(struct page *page,
    >>>> loff_t size;
    >>>> unsigned long len;
    >>>> handle_t *handle = NULL;
    >>>> + ext4_lblk_t block;
    >>>> + loff_t disksize;
    >>>> struct buffer_head *page_bufs;
    >>>> + struct buffer_head *bh, *head;
    >>>> struct inode *inode = page->mapping->host;
    >>>>
    >>>> handle = ext4_journal_current_handle();
    >>>> @@ -1662,6 +1665,38 @@ static int ext4_da_writepage(struct page *page,
    >>>> else
    >>>> ret = block_write_full_page(page, ext4_da_get_block_write, wbc);
    >>>>
    >>>> + if (ret)
    >>>> + return ret;
    >>>> + /*
    >>>> + * When called via shrink_page_list and if we don't have any unmapped
    >>>> + * buffer_head we still could have written some new content in an
    >>>> + * already mapped buffer. That means we need to extent i_disksize here
    >>>> + */
    >>>
    >>> In this case(when extend the file without need block allocation),
    >>> wouldn't make sense to update the i_disksize at write_end() time? So
    >>> that the window of i_size different from i_disksize could be much
    >>> smaller in this case.
    >>>
    >>>
    >>> Something like below? (untested)
    >>
    >> In this case you will have to start a transaction in write_begin . With
    >> the below code transaction is started inside page_lock. Also I don't
    >> think we need needed_blocks credit just 1 should be enough because we
    >> are not doing any block allocation here. We just need to update the
    >> inode block.
    >>
    >>
    >
    > diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
    > index 33f940b..bc925c5 100644
    > --- a/fs/ext4/inode.c
    > +++ b/fs/ext4/inode.c
    > @@ -1770,6 +1770,7 @@ static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
    > struct page *page;
    > pgoff_t index;
    > unsigned from, to;
    > + handle_t *handle;
    > struct inode *inode = mapping->host;
    >
    > index = pos >> PAGE_CACHE_SHIFT;
    > @@ -1777,6 +1778,17 @@ static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
    > to = from + len;
    >
    > retry:
    > + /*
    > + * If we are writing towards the end of an already mapped
    > + * buffer_head, we don't do any block allocation. But we
    > + * need to update i_disksize.
    > + */
    > + handle = ext4_journal_start(inode, 1);
    > + if (IS_ERR(handle)) {
    > + ret = PTR_ERR(handle);
    > + goto out;
    > + }
    > +
    > page = __grab_cache_page(mapping, index);
    > if (!page)
    > return -ENOMEM;
    > @@ -1786,15 +1798,63 @@ static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
    > ext4_da_get_block_prep);
    > if (ret < 0) {
    > unlock_page(page);
    > + ext4_journal_stop(handle);
    > page_cache_release(page);
    > }
    >
    > if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
    > goto retry;
    >
    > +out:
    > return ret;
    > }
    >
    > +static int ext4_da_write_end(struct file *file,
    > + struct address_space *mapping,
    > + loff_t pos, unsigned len, unsigned copied,
    > + struct page *page, void *fsdata)
    > +{
    > + loff_t new_i_size;
    > + unsigned from, to;
    > + int ret = 0, ret2;
    > + struct inode *inode = mapping->host;
    > + handle_t *handle = ext4_journal_current_handle();
    > +
    > + from = pos & (PAGE_CACHE_SIZE - 1);
    > + to = from + len;
    > +
    > + /*
    > + * generic_write_end() will run mark_inode_dirty() if i_size
    > + * changes. So let's piggyback the i_disksize mark_inode_dirty
    > + * into that.
    > + */
    > +
    > + new_i_size = pos + copied;
    > + if (new_i_size > EXT4_I(inode)->i_disksize)
    > + if (!walk_page_buffers(NULL, page_buffers(page),
    > + 0, len, NULL,
    > + ext4_bh_unmapped_or_delay)) {
    > + /*
    > + * Updating i_disksize when extending file without
    > + * need block allocation
    > + */
    > + if (ext4_should_order_data(inode))
    > + ret = ext4_jbd2_file_inode(handle, inode);
    > +
    > + EXT4_I(inode)->i_disksize = new_i_size;
    > + }
    > + ret2 = generic_write_end(file, mapping, pos,
    > + len, copied, page, fsdata);
    > + copied = ret2;
    > + if (ret2 < 0)
    > + ret = ret2;
    > + ret2 = ext4_journal_stop(handle);
    > + if (!ret)
    > + ret = ret2;
    > +
    > + return ret ? ret : copied;
    > +}
    > +
    > static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
    > {
    > /*
    > @@ -2250,7 +2310,7 @@ static int ext4_journalled_set_page_dirty(struct page *page)
    > .writepages = ext4_da_writepages,
    > .sync_page = block_sync_page,
    > .write_begin = ext4_da_write_begin,
    > - .write_end = generic_write_end,
    > + .write_end = ext4_da_write_end,
    > .bmap = ext4_bmap,
    > .invalidatepage = ext4_da_invalidatepage,
    > .releasepage = ext4_releasepage,
    >
    Yes, with this patch applied on top of latest patch queue I no longer
    get truncated files, after running a short test. Tomorrow I will do some
    more thorough testing and use the patch you have send to me in a separate
    mail. The above patch did not apply but it was easy to apply by hand.

    Thanks a lot for the patch!

    Holger



    \
     
     \ /
      Last update: 2008-06-24 23:15    [W:0.036 / U:94.632 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site