The original direct splicing mechanism from Jens required the input to
be a regular file because it was avoiding the special socket case. It
also recognized blkdevs as being close enough to a regular file. But it
forgot about chardevs, which behave the same way and work fine here.

This commit adds the missing S_ISCHR condition so that chardevs such as
/dev/urandom can be directly spliced without strangely returning
-EINVAL.

Cc: Jens Axboe <axboe@suse.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Fixes: b92ce5589374 ("[PATCH] splice: add direct fd <-> fd splicing support")
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
---
 fs/splice.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/splice.c b/fs/splice.c
index 047b79db8eb5..7e673b1786fb 100644
--- a/fs/splice.c
+++ b/fs/splice.c
@@ -824,7 +824,7 @@ ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
 	 * piped splicing for that!
 	 */
 	i_mode = file_inode(in)->i_mode;
-	if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
+	if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISCHR(i_mode)))
 		return -EINVAL;
 
 	/*
-- 
2.35.1

On 5/20/22 3:57 AM, Jason A. Donenfeld wrote:
> The original direct splicing mechanism from Jens required the input to
> be a regular file because it was avoiding the special socket case. It
> also recognized blkdevs as being close enough to a regular file. But it
> forgot about chardevs, which behave the same way and work fine here.
> 
> This commit adds the missing S_ISCHR condition so that chardevs such as
> /dev/urandom can be directly spliced without strangely returning
> -EINVAL.

Should be fine to turn this on for char devices:

Reviewed-by: Jens Axboe <axboe@kernel.dk>

-- 
Jens Axboe

Hey Al,

On 5/20/22, Jens Axboe <axboe@kernel.dk> wrote:
> On 5/20/22 3:57 AM, Jason A. Donenfeld wrote:
>> The original direct splicing mechanism from Jens required the input to
>> be a regular file because it was avoiding the special socket case. It
>> also recognized blkdevs as being close enough to a regular file. But it
>> forgot about chardevs, which behave the same way and work fine here.
>>
>> This commit adds the missing S_ISCHR condition so that chardevs such as
>> /dev/urandom can be directly spliced without strangely returning
>> -EINVAL.
>
> Should be fine to turn this on for char devices:
>
> Reviewed-by: Jens Axboe <axboe@kernel.dk>
>

Was wondering if this would make 5.19. That'd be nice, as it's the
release in which we switch to read_iter().

Jason

Hey again,

On Thu, May 26, 2022 at 11:19:46PM +0200, Jason A. Donenfeld wrote:
> Hey Al,
> 
> On 5/20/22, Jens Axboe <axboe@kernel.dk> wrote:
> > On 5/20/22 3:57 AM, Jason A. Donenfeld wrote:
> >> The original direct splicing mechanism from Jens required the input to
> >> be a regular file because it was avoiding the special socket case. It
> >> also recognized blkdevs as being close enough to a regular file. But it
> >> forgot about chardevs, which behave the same way and work fine here.
> >>
> >> This commit adds the missing S_ISCHR condition so that chardevs such as
> >> /dev/urandom can be directly spliced without strangely returning
> >> -EINVAL.
> >
> > Should be fine to turn this on for char devices:
> >
> > Reviewed-by: Jens Axboe <axboe@kernel.dk>
> >
> 
> Was wondering if this would make 5.19. That'd be nice, as it's the
> release in which we switch to read_iter().

Just thought I should ping once more on this. Should probably be queued
up somewhat soon for 5.19 if it's to make 5.19, which I would really
appreciate.

Jason