gfs2: use iomap for buffered I/O in ordered and writeback mode
Subsystems (2)
Andreas Gruenbacher
maintainerHi Christoph,
Quoted 4 lines · Andreas Gruenbacher
Quoted 1 line · Christoph Hellwig
On Mon, Aug 05, 2019 at 02:27:21PM +0200, Andreas Gruenbacher wrote:
here are the changes we currently need on top of what you've posted on July 1. [...]
again, thank you for this patch. After fixing some related bugs around this change, it seems I've finally got this to work properly. Below are the minor changes I needed to make on top of your version.
This requires functions iomap_page_create and iomap_set_range_uptodate to be exported; i'll post a patch for that sepatately.
The result can be found here:
git://git.kernel.org/pub/scm/linux/kernel/git/gfs2/linux-gfs2.git for-next.iomap
Thanks,
Andreas
--- fs/gfs2/bmap.c | 6 ++++-- fs/gfs2/file.c | 2 ++ 2 files changed, 6 insertions(+), 2 deletions(-)
fs/gfs2/bmap.c +4 -2
@@ -75,13 +75,12 @@ static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,memcpy (kaddr, dibh->b_data +sizeof (struct gfs2_dinode ), dsize);memset (kaddr + dsize,0 ,PAGE_SIZE - dsize);kunmap (page); - -SetPageUptodate (page); }if (gfs2_is_jdata (ip)) {struct buffer_head *bh; +SetPageUptodate (page);if (!page_has_buffers (page))create_empty_buffers (page,BIT (inode->i_blkbits),BIT (BH_Uptodate));
@@ -93,6 +92,9 @@ static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,set_buffer_uptodate (bh);gfs2_trans_add_data (ip->i_gl, bh); }else { +iomap_page_create (inode, page); +iomap_set_range_uptodate (page,0 ,i_blocksize (inode)); +set_page_dirty (page);gfs2_ordered_add_inode (ip); }
fs/gfs2/file.c +2 -0
@@ -555,6 +555,8 @@ static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf) out_uninit:gfs2_holder_uninit (&gh);if (ret ==0 ) { +if (!gfs2_is_jdata (ip)) +iomap_page_create (inode, page);set_page_dirty (page);wait_for_stable_page (page); }
Signature · 1 line
2.20.1
Quoted 16 lines · Andreas Gruenbacher
fs/gfs2/bmap.c +1
-2@@ -75,13 +75,12 @@ static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,memcpy (kaddr, dibh->b_data +sizeof (struct gfs2_dinode ), dsize);memset (kaddr + dsize,0 ,PAGE_SIZE - dsize);kunmap (page); - -SetPageUptodate (page); }if (gfs2_is_jdata (ip)) {struct buffer_head *bh; +SetPageUptodate (page);if (!page_has_buffers (page))create_empty_buffers (page,BIT (inode->i_blkbits),BIT (BH_Uptodate));
fs/gfs2/bmap.c static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh, · line 99
Quoted 8 lines
fs/gfs2/bmap.c +3
-0@@ -93,6 +92,9 @@ static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,set_buffer_uptodate (bh);gfs2_trans_add_data (ip->i_gl, bh); }else { +iomap_page_create (inode, page); +iomap_set_range_uptodate (page,0 ,i_blocksize (inode)); +set_page_dirty (page);gfs2_ordered_add_inode (ip); }
Can you create a helper that copies the data from a passed in kernel pointer, length pair into the page, then marks it uptodate and dirty, please?
fs/gfs2/file.c static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf) · line 559
Quoted 6 lines
fs/gfs2/file.c +2
-0@@ -555,6 +555,8 @@ static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf) out_uninit:gfs2_holder_uninit (&gh);if (ret ==0 ) { +if (!gfs2_is_jdata (ip)) +iomap_page_create (inode, page);
What is this one for? The iomap_page is supposed to use lazy allocation, that is we only allocate it once it is used. What code expects the structure but doesn't see it without this hunk? I guess it is iomap_writepage_map, which should probably just switch to call iomap_page_create.
That being said is there any way we can get gfs2 to use iomap_page_mkwrite for the !jdata case?