Skip to contentSkip to conversation

splice: allow non-block splice only when in file is seekable

PATCH Superseded Changli Gao 2 participants 4 messages May 26, 2010 VFS

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 2/3 splice: allow non-block splice only when in file is seekable

allow non-block splice only when in file is seekable.

do_splice_to() is split to two parts: read data from in file to spd, and move data from spd to pipe. If there isn't much space in pipe for the data in spd, and the splice is called with non-block flag, the data dropped due to non enough space in pipe will be lost forever when in file isn't seekable.

----

 fs/splice.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
fs/splice.c +5 -1
index 9e52de5..8137e23 100644
@@ -1346,8 +1346,12 @@ static long do_splice(struct file *in, loff_t __user *off_in,
 			if (copy_from_user(&offset, off_in, sizeof(loff_t)))
 				return -EFAULT;
 			off = &offset;
-		} else
+		} else {
+			if ((flags & SPLICE_F_NONBLOCK) &&
+			    !(in->f_mode & FMODE_PREAD))
+				return -EINVAL;
 			off = &in->f_pos;
+		}
 
 		ret = do_splice_to(in, off, opipe, len, flags);
 
#1 lore raw
Quoted 6 lines · Changli Gao

allow non-block splice only when in file is seekable.

do_splice_to() is split to two parts: read data from in file to spd, and move data from spd to pipe. If there isn't much space in pipe for the data in spd, and the splice is called with non-block flag, the data dropped due to non enough space in pipe will be lost forever when in file isn't seekable.

This is a gratuitous ABI change. I don't think we should hold the hand of application writers, maybe they _know_ the splice will not block.

NACK from me.

Miklos

Quoted 28 lines

----

 fs/splice.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
fs/splice.c +5 -2
index 9e52de5..8137e23 100644
@@ -1346,8 +1346,12 @@ static long do_splice(struct file *in, loff_t __user *off_in,
 			if (copy_from_user(&offset, off_in, sizeof(loff_t)))
 				return -EFAULT;
 			off = &offset;
-		} else
+		} else {
+			if ((flags & SPLICE_F_NONBLOCK) &&
+			    !(in->f_mode & FMODE_PREAD))
+				return -EINVAL;
 			off = &in->f_pos;
+		}
 
 		ret = do_splice_to(in, off, opipe, len, flags);
 
--

To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#2 lore raw
Quoted 10 lines · Miklos Szeredi
Quoted 6 lines · Changli Gao

allow non-block splice only when in file is seekable.

do_splice_to() is split to two parts: read data from in file to spd, and move data from spd to pipe. If there isn't much space in pipe for the data in spd, and the splice is called with non-block flag, the data dropped due to non enough space in pipe will be lost forever when in file isn't seekable.

This is a gratuitous ABI change.  I don't think we should hold the hand of application writers,

Yea, it is an ABI change. But without that, we'll fail user silently currently. Some fundamental change is needed to fix this issue fully, if you think it is a bug.

Quoted 3 lines

maybe they _know_ the splice will not
block.

If they know the splice won't block, why he specifies a non-block flag?

Signature · 2 lines
Regards,
Changli Gao([email protected])
#3 lore raw
Quoted 9 lines · Changli Gao
Quoted 8 lines · Miklos Szeredi
Quoted 6 lines · Changli Gao

allow non-block splice only when in file is seekable.

do_splice_to() is split to two parts: read data from in file to spd, and move data from spd to pipe. If there isn't much space in pipe for the data in spd, and the splice is called with non-block flag, the data dropped due to non enough space in pipe will be lost forever when in file isn't seekable.

Quoted 6 lines
Quoted 2 lines

This is a gratuitous ABI change.  I don't think we should hold the hand of application writers,

Yea, it is an ABI change. But without that, we'll fail user silently currently. Some fundamental change is needed to fix this issue fully, if you think it is a bug.

I don't think it's a bug. The same thing will happen without SPLICE_F_NONBLOCK if splice_to_pipe() is interrupted by a signal while it blocks.

To prevent data loss the application needs to be aware of the fact that doing a splice from a non-seekable fd needs to be done carefully, making sure that it doesn't block on the pipe.

Miklos
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html