Fix the new filemap_splice_read() function to get i_size from
in->f_mapping->host, not in->f_inode so that it works with block devices
too (in->f_inode points to the device file, which is typically zero size).

Fixes: 07073eb01c5f ("splice: Add a func to do a splice from a buffered file without ITER_PIPE")
Link: https://lore.kernel.org/r/0c6b661c-f7ff-cf12-b7f0-00b6b2f1317b@amd.com/
Reported-by: Ayush Jain <ayush.jain3@amd.com>
cc: Jens Axboe <axboe@kernel.dk>
cc: Christoph Hellwig <hch@lst.de>
cc: Al Viro <viro@zeniv.linux.org.uk>
cc: David Hildenbrand <david@redhat.com>
cc: John Hubbard <jhubbard@nvidia.com>
cc: Steve French <stfrench@microsoft.com>
cc: linux-mm@kvack.org
cc: linux-block@vger.kernel.org
cc: linux-fsdevel@vger.kernel.org
---
 mm/filemap.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/mm/filemap.c b/mm/filemap.c
index 470be06b6096..f86cc8acf33a 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -2902,7 +2902,7 @@ ssize_t filemap_splice_read(struct file *in, loff_t *ppos,
 	do {
 		cond_resched();
 
-		if (*ppos >= i_size_read(file_inode(in)))
+		if (*ppos >= i_size_read(in->f_mapping->host))
 			break;
 
 		iocb.ki_pos = *ppos;
@@ -2918,7 +2918,7 @@ ssize_t filemap_splice_read(struct file *in, loff_t *ppos,
 		 * part of the page is not copied back to userspace (unless
 		 * another truncate extends the file - this is desired though).
 		 */
-		isize = i_size_read(file_inode(in));
+		isize = i_size_read(in->f_mapping->host);
 		if (unlikely(*ppos >= isize))
 			break;
 		end_offset = min_t(loff_t, isize, *ppos + len);

On Wed, 19 Apr 2023 09:54:11 +0100, David Howells wrote:
> Fix the new filemap_splice_read() function to get i_size from
> in->f_mapping->host, not in->f_inode so that it works with block devices
> too (in->f_inode points to the device file, which is typically zero size).
> 
> 

Applied, thanks!

[1/1] splice: Fix filemap of a blockdev
      commit: 5a9515a407d1aec1dd76c24fe99d9981730b74fb

Best regards,
-- 
Jens Axboe

Note that this shouldn't affect direct_splice_read() as that doesn't look at
the size of the file, but rather just keeps reading from it until the
requested amount is read, the pipe is full or ->read_iter() indicates EOF by
returning 0 (so it could work for copy-splicing from a pipe/socket/chardev
too).

David

Jens Axboe <axboe@kernel.dk> wrote:

> [1/1] splice: Fix filemap of a blockdev

Actually, would you be able to fix the subject?  I left a word out:

	splice: Fix filemap splice of a blockdev

or maybe:

	splice: Fix buffered splice of a blockdev

Sorry about that,
David

On 4/19/23 6:57 AM, David Howells wrote:
> Jens Axboe <axboe@kernel.dk> wrote:
> 
>> [1/1] splice: Fix filemap of a blockdev
> 
> Actually, would you be able to fix the subject?  I left a word out:
> 
> 	splice: Fix filemap splice of a blockdev
> 
> or maybe:
> 
> 	splice: Fix buffered splice of a blockdev

Fixed it up.

-- 
Jens Axboe

Hello David,

This resolves the Fio-test issue that i reported in
Link: https://lore.kernel.org/r/0c6b661c-f7ff-cf12-b7f0-00b6b2f1317b@amd.com/

On 4/19/2023 2:24 PM, David Howells wrote:
> Fix the new filemap_splice_read() function to get i_size from
> in->f_mapping->host, not in->f_inode so that it works with block devices
> too (in->f_inode points to the device file, which is typically zero size).
> 
> Fixes: 07073eb01c5f ("splice: Add a func to do a splice from a buffered file without ITER_PIPE")
> Link: https://lore.kernel.org/r/0c6b661c-f7ff-cf12-b7f0-00b6b2f1317b@amd.com/
> Reported-by: Ayush Jain <ayush.jain3@amd.com>

Tested-by: Ayush Jain <ayush.jain3@amd.com>

> cc: Jens Axboe <axboe@kernel.dk>
> cc: Christoph Hellwig <hch@lst.de>
> cc: Al Viro <viro@zeniv.linux.org.uk>
> cc: David Hildenbrand <david@redhat.com>
> cc: John Hubbard <jhubbard@nvidia.com>
> cc: Steve French <stfrench@microsoft.com>
> cc: linux-mm@kvack.org
> cc: linux-block@vger.kernel.org
> cc: linux-fsdevel@vger.kernel.org
> ---
>   mm/filemap.c |    4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/mm/filemap.c b/mm/filemap.c
> index 470be06b6096..f86cc8acf33a 100644
> --- a/mm/filemap.c
> +++ b/mm/filemap.c
> @@ -2902,7 +2902,7 @@ ssize_t filemap_splice_read(struct file *in, loff_t *ppos,
>   	do {
>   		cond_resched();
>   
> -		if (*ppos >= i_size_read(file_inode(in)))
> +		if (*ppos >= i_size_read(in->f_mapping->host))
>   			break;
>   
>   		iocb.ki_pos = *ppos;
> @@ -2918,7 +2918,7 @@ ssize_t filemap_splice_read(struct file *in, loff_t *ppos,
>   		 * part of the page is not copied back to userspace (unless
>   		 * another truncate extends the file - this is desired though).
>   		 */
> -		isize = i_size_read(file_inode(in));
> +		isize = i_size_read(in->f_mapping->host);
>   		if (unlikely(*ppos >= isize))
>   			break;
>   		end_offset = min_t(loff_t, isize, *ppos + len);
> 

Regards,
Ayush Jain