[PATCH v2] Btrfs: add readahead for send_write

[PATCH v2] Btrfs: add readahead for send_write

[Liu Bo]

Btrfs send reads data from disk and then writes to a stream via pipe or
a file via flush.

Currently we're going to read each page at a time, so every page results
in a disk read, which is not friendly to disks, esp. HDD.  Given that,
the performance can be gained by adding readahead for those pages.

Here is a quick test:
$ btrfs subvolume create send
$ xfs_io -f -c "pwrite 0 1G" send/foobar
$ btrfs subvolume snap -r send ro
$ time "btrfs send ro -f /dev/null"

           w/o             w
real    1m37.527s       0m9.097s
user    0m0.122s        0m0.086s
sys     0m53.191s       0m12.857s

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
---
v1->v2: return ENOMEM on failing to allocate memory.

 fs/btrfs/send.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/fs/btrfs/send.c b/fs/btrfs/send.c
index 9dde971..d55faa7 100644
--- a/fs/btrfs/send.c
+++ b/fs/btrfs/send.c
@@ -3972,6 +3972,7 @@ static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
 	pgoff_t last_index;
 	unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
 	ssize_t ret = 0;
+	struct file_ra_state *ra = NULL;
 
 	key.objectid = sctx->cur_ino;
 	key.type = BTRFS_INODE_ITEM_KEY;
@@ -3991,6 +3992,17 @@ static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
 		goto out;
 
 	last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
+
+	/* initial readahead */
+	ra = kzalloc(sizeof(*ra), GFP_NOFS);
+	if (!ra) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	file_ra_state_init(ra, inode->i_mapping);
+	btrfs_force_ra(inode->i_mapping, ra, NULL, index, last_index-index + 1);
+
 	while (index <= last_index) {
 		unsigned cur_len = min_t(unsigned, len,
 					 PAGE_CACHE_SIZE - pg_offset);
@@ -4022,6 +4034,7 @@ static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
 		ret += cur_len;
 	}
 out:
+	kfree(ra);
 	iput(inode);
 	return ret;
 }
-- 
1.8.1.4

Re: [PATCH v2] Btrfs: add readahead for send_write

[David Sterba]

On Mon, Mar 03, 2014 at 05:24:43PM +0800, Liu Bo wrote:
> --- a/fs/btrfs/send.c
> +++ b/fs/btrfs/send.c
> @@ -3972,6 +3972,7 @@ static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
>  	pgoff_t last_index;
>  	unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
>  	ssize_t ret = 0;
> +	struct file_ra_state *ra = NULL;
>  
>  	key.objectid = sctx->cur_ino;
>  	key.type = BTRFS_INODE_ITEM_KEY;
> @@ -3991,6 +3992,17 @@ static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
>  		goto out;
>  
>  	last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
> +
> +	/* initial readahead */
> +	ra = kzalloc(sizeof(*ra), GFP_NOFS);
> +	if (!ra) {
> +		ret = -ENOMEM;

This should not be a hard failure, it can continue without RA for this
buffer. Besides, the RA buffer can be allocated at the beginning of send
operation and pointer stored in the send context.

> +		goto out;
> +	}
> +
> +	file_ra_state_init(ra, inode->i_mapping);
> +	btrfs_force_ra(inode->i_mapping, ra, NULL, index, last_index-index + 1);
> +
>  	while (index <= last_index) {
>  		unsigned cur_len = min_t(unsigned, len,
>  					 PAGE_CACHE_SIZE - pg_offset);
> @@ -4022,6 +4034,7 @@ static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
>  		ret += cur_len;
>  	}
>  out:
> +	kfree(ra);
>  	iput(inode);
>  	return ret;
>  }
> -- 
> 1.8.1.4
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2] Btrfs: add readahead for send_write

[Liu Bo]

On Mon, Mar 03, 2014 at 03:25:28PM +0100, David Sterba wrote:
> On Mon, Mar 03, 2014 at 05:24:43PM +0800, Liu Bo wrote:
> > --- a/fs/btrfs/send.c
> > +++ b/fs/btrfs/send.c
> > @@ -3972,6 +3972,7 @@ static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
> >  	pgoff_t last_index;
> >  	unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
> >  	ssize_t ret = 0;
> > +	struct file_ra_state *ra = NULL;
> >  
> >  	key.objectid = sctx->cur_ino;
> >  	key.type = BTRFS_INODE_ITEM_KEY;
> > @@ -3991,6 +3992,17 @@ static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
> >  		goto out;
> >  
> >  	last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
> > +
> > +	/* initial readahead */
> > +	ra = kzalloc(sizeof(*ra), GFP_NOFS);
> > +	if (!ra) {
> > +		ret = -ENOMEM;
> 
> This should not be a hard failure, it can continue without RA for this
> buffer. Besides, the RA buffer can be allocated at the beginning of send
> operation and pointer stored in the send context.

Good point, I think we can fold this into 'struct send_ctx', do you agree?

Thanks,
-liubo

Re: [PATCH v2] Btrfs: add readahead for send_write

[David Sterba]

On Tue, Mar 04, 2014 at 10:49:52AM +0800, Liu Bo wrote:
> On Mon, Mar 03, 2014 at 03:25:28PM +0100, David Sterba wrote:
> > This should not be a hard failure, it can continue without RA for this
> > buffer. Besides, the RA buffer can be allocated at the beginning of send
> > operation and pointer stored in the send context.
> 
> Good point, I think we can fold this into 'struct send_ctx', do you agree?

Sure, size of send_ctx + readahead state will remain below 512.