[PATCH v2 3/4]: sendfile: remove flags paramter of do_splice_direct()

[PATCH v2 3/4]: sendfile: remove flags paramter of do_splice_direct()

[Changli Gao]

remove flags paramter of do_splice_direct().

the flags parameter of do_splice_direct() doesn't mean non-block read of
in file, but non-block read/write of pipe. As the reader and writer of process
internal pipe is the process itself, we don't need this flags.

Signed-off-by: Changli Gao <xiaosuo@gmail.com>
----
 fs/read_write.c    |   15 ++-------------
 fs/splice.c        |   19 +++----------------
 include/linux/fs.h |    2 +-
 3 files changed, 6 insertions(+), 30 deletions(-)
diff --git a/fs/read_write.c b/fs/read_write.c
index 113386d..0688e05 100644
--- a/fs/read_write.c
+++ b/fs/read_write.c
@@ -796,7 +796,7 @@ static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
 	struct inode * in_inode, * out_inode;
 	loff_t pos;
 	ssize_t retval;
-	int fput_needed_in, fput_needed_out, fl;
+	int fput_needed_in, fput_needed_out;
 
 	/*
 	 * Get input file, and verify that it is ok..
@@ -846,18 +846,7 @@ static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
 		count = max - pos;
 	}
 
-	fl = 0;
-#if 0
-	/*
-	 * We need to debate whether we can enable this or not. The
-	 * man page documents EAGAIN return for the output at least,
-	 * and the application is arguably buggy if it doesn't expect
-	 * EAGAIN on a non-blocking file descriptor.
-	 */
-	if (in_file->f_flags & O_NONBLOCK)
-		fl = SPLICE_F_NONBLOCK;
-#endif
-	retval = do_splice_direct(in_file, ppos, out_file, count, fl);
+	retval = do_splice_direct(in_file, ppos, out_file, count);
 
 	if (retval > 0) {
 		add_rchar(current, retval);
diff --git a/fs/splice.c b/fs/splice.c
index eb602cb..57172ca 100644
--- a/fs/splice.c
+++ b/fs/splice.c
@@ -1125,7 +1125,7 @@ ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
 	long ret, bytes;
 	umode_t i_mode;
 	size_t len;
-	int i, flags;
+	int i;
 
 	/*
 	 * We require the input being a regular file, as we don't want to
@@ -1162,29 +1162,18 @@ ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
 	ret = 0;
 	bytes = 0;
 	len = sd->total_len;
-	flags = sd->flags;
-
-	/*
-	 * Don't block on output, we have to drain the direct pipe.
-	 */
-	sd->flags &= ~SPLICE_F_NONBLOCK;
 
 	while (len) {
 		size_t read_len;
 		loff_t pos = sd->pos, prev_pos = pos;
 
-		ret = do_splice_to(in, &pos, pipe, len, flags);
+		ret = do_splice_to(in, &pos, pipe, len, 0);
 		if (unlikely(ret <= 0))
 			goto out_release;
 
 		read_len = ret;
 		sd->total_len = read_len;
 
-		/*
-		 * NOTE: nonblocking mode only applies to the input. We
-		 * must not do the output in nonblocking mode as then we
-		 * could get stuck data in the internal pipe:
-		 */
 		ret = actor(pipe, sd);
 		if (unlikely(ret <= 0)) {
 			sd->pos = prev_pos;
@@ -1242,7 +1231,6 @@ static int direct_splice_actor(struct pipe_inode_info *pipe,
  * @ppos:	input file offset
  * @out:	file to splice to
  * @len:	number of bytes to splice
- * @flags:	splice modifier flags
  *
  * Description:
  *    For use by do_sendfile(). splice can easily emulate sendfile, but
@@ -1252,12 +1240,11 @@ static int direct_splice_actor(struct pipe_inode_info *pipe,
  *
  */
 long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
-		      size_t len, unsigned int flags)
+		      size_t len)
 {
 	struct splice_desc sd = {
 		.len		= len,
 		.total_len	= len,
-		.flags		= flags,
 		.pos		= *ppos,
 		.u.file		= out,
 	};
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 4079ef9..eb86433 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -2224,7 +2224,7 @@ extern ssize_t generic_file_splice_write(struct pipe_inode_info *,
 extern ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe,
 		struct file *out, loff_t *, size_t len, unsigned int flags);
 extern long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
-		size_t len, unsigned int flags);
+		size_t len);
 
 extern void
 file_ra_state_init(struct file_ra_state *ra, struct address_space *mapping);

Re: [PATCH v2 3/4]: sendfile: remove flags paramter of do_splice_direct()

[Miklos Szeredi]

On Wed, 26 May 2010, Changli Gao wrote:
> remove flags paramter of do_splice_direct().
> 
> the flags parameter of do_splice_direct() doesn't mean non-block read of
> in file,

Actually, it does.  Look at the SPLICE_F_NONBLOCK usage in
__generic_splice_file_read().

>  but non-block read/write of pipe. As the reader and writer of process
> internal pipe is the process itself, we don't need this flags.
> 
> Signed-off-by: Changli Gao <xiaosuo@gmail.com>
> ----
>  fs/read_write.c    |   15 ++-------------
>  fs/splice.c        |   19 +++----------------
>  include/linux/fs.h |    2 +-
>  3 files changed, 6 insertions(+), 30 deletions(-)
> diff --git a/fs/read_write.c b/fs/read_write.c
> index 113386d..0688e05 100644
> --- a/fs/read_write.c
> +++ b/fs/read_write.c
> @@ -796,7 +796,7 @@ static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
>  	struct inode * in_inode, * out_inode;
>  	loff_t pos;
>  	ssize_t retval;
> -	int fput_needed_in, fput_needed_out, fl;
> +	int fput_needed_in, fput_needed_out;
>  
>  	/*
>  	 * Get input file, and verify that it is ok..
> @@ -846,18 +846,7 @@ static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
>  		count = max - pos;
>  	}
>  
> -	fl = 0;
> -#if 0
> -	/*
> -	 * We need to debate whether we can enable this or not. The
> -	 * man page documents EAGAIN return for the output at least,
> -	 * and the application is arguably buggy if it doesn't expect
> -	 * EAGAIN on a non-blocking file descriptor.
> -	 */
> -	if (in_file->f_flags & O_NONBLOCK)
> -		fl = SPLICE_F_NONBLOCK;
> -#endif
> -	retval = do_splice_direct(in_file, ppos, out_file, count, fl);
> +	retval = do_splice_direct(in_file, ppos, out_file, count);
>  
>  	if (retval > 0) {
>  		add_rchar(current, retval);
> diff --git a/fs/splice.c b/fs/splice.c
> index eb602cb..57172ca 100644
> --- a/fs/splice.c
> +++ b/fs/splice.c
> @@ -1125,7 +1125,7 @@ ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
>  	long ret, bytes;
>  	umode_t i_mode;
>  	size_t len;
> -	int i, flags;
> +	int i;
>  
>  	/*
>  	 * We require the input being a regular file, as we don't want to
> @@ -1162,29 +1162,18 @@ ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
>  	ret = 0;
>  	bytes = 0;
>  	len = sd->total_len;
> -	flags = sd->flags;
> -
> -	/*
> -	 * Don't block on output, we have to drain the direct pipe.
> -	 */
> -	sd->flags &= ~SPLICE_F_NONBLOCK;
>  
>  	while (len) {
>  		size_t read_len;
>  		loff_t pos = sd->pos, prev_pos = pos;
>  
> -		ret = do_splice_to(in, &pos, pipe, len, flags);
> +		ret = do_splice_to(in, &pos, pipe, len, 0);
>  		if (unlikely(ret <= 0))
>  			goto out_release;
>  
>  		read_len = ret;
>  		sd->total_len = read_len;
>  
> -		/*
> -		 * NOTE: nonblocking mode only applies to the input. We
> -		 * must not do the output in nonblocking mode as then we
> -		 * could get stuck data in the internal pipe:
> -		 */
>  		ret = actor(pipe, sd);
>  		if (unlikely(ret <= 0)) {
>  			sd->pos = prev_pos;
> @@ -1242,7 +1231,6 @@ static int direct_splice_actor(struct pipe_inode_info *pipe,
>   * @ppos:	input file offset
>   * @out:	file to splice to
>   * @len:	number of bytes to splice
> - * @flags:	splice modifier flags
>   *
>   * Description:
>   *    For use by do_sendfile(). splice can easily emulate sendfile, but
> @@ -1252,12 +1240,11 @@ static int direct_splice_actor(struct pipe_inode_info *pipe,
>   *
>   */
>  long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
> -		      size_t len, unsigned int flags)
> +		      size_t len)
>  {
>  	struct splice_desc sd = {
>  		.len		= len,
>  		.total_len	= len,
> -		.flags		= flags,
>  		.pos		= *ppos,
>  		.u.file		= out,
>  	};
> diff --git a/include/linux/fs.h b/include/linux/fs.h
> index 4079ef9..eb86433 100644
> --- a/include/linux/fs.h
> +++ b/include/linux/fs.h
> @@ -2224,7 +2224,7 @@ extern ssize_t generic_file_splice_write(struct pipe_inode_info *,
>  extern ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe,
>  		struct file *out, loff_t *, size_t len, unsigned int flags);
>  extern long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
> -		size_t len, unsigned int flags);
> +		size_t len);
>  
>  extern void
>  file_ra_state_init(struct file_ra_state *ra, struct address_space *mapping);
> --
> To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" 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 3/4]: sendfile: remove flags paramter of do_splice_direct()

[Changli Gao]

On Fri, May 28, 2010 at 5:40 PM, Miklos Szeredi <miklos@szeredi.hu> wrote:
> On Wed, 26 May 2010, Changli Gao wrote:
>> remove flags paramter of do_splice_direct().
>>
>> the flags parameter of do_splice_direct() doesn't mean non-block read of
>> in file,
>
> Actually, it does.  Look at the SPLICE_F_NONBLOCK usage in
> __generic_splice_file_read().
>

Oh, I checked the code again. You're right. However, why don't we
check the flags of in file instead in __generic_file_splice_read()?

-- 
Regards，
Changli Gao(xiaosuo@gmail.com)

Re: [PATCH v2 3/4]: sendfile: remove flags paramter of do_splice_direct()

[Changli Gao]

On Fri, May 28, 2010 at 6:06 PM, Changli Gao <xiaosuo@gmail.com> wrote:
> On Fri, May 28, 2010 at 5:40 PM, Miklos Szeredi <miklos@szeredi.hu> wrote:
>> On Wed, 26 May 2010, Changli Gao wrote:
>>> remove flags paramter of do_splice_direct().
>>>
>>> the flags parameter of do_splice_direct() doesn't mean non-block read of
>>> in file,
>>
>> Actually, it does.  Look at the SPLICE_F_NONBLOCK usage in
>> __generic_splice_file_read().
>>
>
> Oh, I checked the code again. You're right. However, why don't we
> check the flags of in file instead in __generic_file_splice_read()?
>

I am afraid that __generic_file_splice_read() should not use SPLICE_F_NONBLOCK.

Here is the comment from the commit 42324c62704365d6a3e89138dea55909d2f26afe :

<<EOF
Linus introduced  SPLICE_F_NONBLOCK in commit
29e350944fdc2dfca102500790d8ad6d6ff4f69d
(splice: add SPLICE_F_NONBLOCK flag )

  It doesn't make the splice itself necessarily nonblocking (because the
  actual file descriptors that are spliced from/to may block unless they
  have the O_NONBLOCK flag set), but it makes the splice pipe operations
  nonblocking.

Linus intention was clear : let SPLICE_F_NONBLOCK control the splice
pipe mode only
EOF

And I have greped the whole code:

fs/splice.c:            if (spd->flags & SPLICE_F_NONBLOCK) {
fs/splice.c:                    if (flags & SPLICE_F_NONBLOCK) {
fs/splice.c:            if (sd->flags & SPLICE_F_NONBLOCK)
fs/splice.c:                    if (flags & SPLICE_F_NONBLOCK) {
fs/splice.c:            if (flags & SPLICE_F_NONBLOCK) {
fs/splice.c:                    if (flags & SPLICE_F_NONBLOCK) {
fs/splice.c:    if (!ret && ipipe->waiting_writers && (flags &
SPLICE_F_NONBLOCK))
fs/splice.c: * applicable one is SPLICE_F_NONBLOCK.
include/linux/splice.h:#define SPLICE_F_NONBLOCK (0x02) /* don't block
on the pipe splicing (but */
kernel/relay.c:                 if (flags & SPLICE_F_NONBLOCK)
kernel/trace/trace.c:           if (flags & SPLICE_F_NONBLOCK)

besides the __generic_file_splice_read(), there are two users: relay
and trace. I believe they all misuse this flag.

-- 
Regards，
Changli Gao(xiaosuo@gmail.com)
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" 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 3/4]: sendfile: remove flags paramter of do_splice_direct()

[Steven Rostedt]

[ Removed matthew@wil.cx since I'm getting errors in sending to that address ]


On Sat, 2010-05-29 at 09:59 +0800, Changli Gao wrote:
> On Fri, May 28, 2010 at 6:06 PM, Changli Gao <xiaosuo@gmail.com> wrote:
> > On Fri, May 28, 2010 at 5:40 PM, Miklos Szeredi <miklos@szeredi.hu> wrote:
> >> On Wed, 26 May 2010, Changli Gao wrote:
> >>> remove flags paramter of do_splice_direct().
> >>>
> >>> the flags parameter of do_splice_direct() doesn't mean non-block read of
> >>> in file,
> >>
> >> Actually, it does.  Look at the SPLICE_F_NONBLOCK usage in
> >> __generic_splice_file_read().
> >>
> >
> > Oh, I checked the code again. You're right. However, why don't we
> > check the flags of in file instead in __generic_file_splice_read()?
> >
> 
> I am afraid that __generic_file_splice_read() should not use SPLICE_F_NONBLOCK.
> 
> Here is the comment from the commit 42324c62704365d6a3e89138dea55909d2f26afe :
> 
> <<EOF
> Linus introduced  SPLICE_F_NONBLOCK in commit
> 29e350944fdc2dfca102500790d8ad6d6ff4f69d
> (splice: add SPLICE_F_NONBLOCK flag )
> 
>   It doesn't make the splice itself necessarily nonblocking (because the
>   actual file descriptors that are spliced from/to may block unless they
>   have the O_NONBLOCK flag set), but it makes the splice pipe operations
>   nonblocking.
> 
> Linus intention was clear : let SPLICE_F_NONBLOCK control the splice
> pipe mode only
> EOF
> 
> And I have greped the whole code:
> 


> kernel/trace/trace.c:           if (flags & SPLICE_F_NONBLOCK)
> 
> besides the __generic_file_splice_read(), there are two users: relay
> and trace. I believe they all misuse this flag.
> 


The trace splice never blocks anyway. Since we have no good method to
wake up a waiter. That is, we never know if it is safe to call wakeup,
because a tracepoint may be in a section where the rq lock is held.

The only thing that passing SPLICE_F_NONBLOCK does in that code is to
return -EAGAIN when there is nothing in the buffer. Otherwise, it
returns 0.

-- Steve

Re: [PATCH v2 3/4]: sendfile: remove flags paramter of do_splice_direct()

[Changli Gao]

On Sun, May 30, 2010 at 12:20 AM, Steven Rostedt <rostedt@goodmis.org> wrote:
> [ Removed matthew@wil.cx since I'm getting errors in sending to that address ]
>
>
> On Sat, 2010-05-29 at 09:59 +0800, Changli Gao wrote:
>> On Fri, May 28, 2010 at 6:06 PM, Changli Gao <xiaosuo@gmail.com> wrote:
>> > On Fri, May 28, 2010 at 5:40 PM, Miklos Szeredi <miklos@szeredi.hu> wrote:
>> >> On Wed, 26 May 2010, Changli Gao wrote:
>> >>> remove flags paramter of do_splice_direct().
>> >>>
>> >>> the flags parameter of do_splice_direct() doesn't mean non-block read of
>> >>> in file,
>> >>
>> >> Actually, it does.  Look at the SPLICE_F_NONBLOCK usage in
>> >> __generic_splice_file_read().
>> >>
>> >
>> > Oh, I checked the code again. You're right. However, why don't we
>> > check the flags of in file instead in __generic_file_splice_read()?
>> >
>>
>> I am afraid that __generic_file_splice_read() should not use SPLICE_F_NONBLOCK.
>>
>> Here is the comment from the commit 42324c62704365d6a3e89138dea55909d2f26afe :
>>
>> <<EOF
>> Linus introduced  SPLICE_F_NONBLOCK in commit
>> 29e350944fdc2dfca102500790d8ad6d6ff4f69d
>> (splice: add SPLICE_F_NONBLOCK flag )
>>
>>   It doesn't make the splice itself necessarily nonblocking (because the
>>   actual file descriptors that are spliced from/to may block unless they
>>   have the O_NONBLOCK flag set), but it makes the splice pipe operations
>>   nonblocking.
>>
>> Linus intention was clear : let SPLICE_F_NONBLOCK control the splice
>> pipe mode only
>> EOF
>>
>> And I have greped the whole code:
>>
>
>
>> kernel/trace/trace.c:           if (flags & SPLICE_F_NONBLOCK)
>>
>> besides the __generic_file_splice_read(), there are two users: relay
>> and trace. I believe they all misuse this flag.
>>
>
>
> The trace splice never blocks anyway. Since we have no good method to
> wake up a waiter. That is, we never know if it is safe to call wakeup,
> because a tracepoint may be in a section where the rq lock is held.
>
> The only thing that passing SPLICE_F_NONBLOCK does in that code is to
> return -EAGAIN when there is nothing in the buffer. Otherwise, it
> returns 0.
>

But why not use (file->f_flags & O_NONBLOCK) instead, as
tracing_splice_read_pipe() does?


-- 
Regards，
Changli Gao(xiaosuo@gmail.com)
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" 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 3/4]: sendfile: remove flags paramter of do_splice_direct()

[Steven Rostedt]

On Sun, 2010-05-30 at 07:58 +0800, Changli Gao wrote:
> On Sun, May 30, 2010 at 12:20 AM, Steven Rostedt <rostedt@goodmis.org> wrote:
> > [ Removed matthew@wil.cx since I'm getting errors in sending to that address ]
> >

> But why not use (file->f_flags & O_NONBLOCK) instead, as
> tracing_splice_read_pipe() does?

Probably because I copied the code from something that used
SPLICE_F_NONBLOCK ;-)

But userspace passes that flag too. Although, maybe I just don't
understand the API fully.

-- Steve

Re: [PATCH v2 3/4]: sendfile: remove flags paramter of do_splice_direct()

[Changli Gao]

On Sun, May 30, 2010 at 8:39 AM, Steven Rostedt <rostedt@goodmis.org> wrote:
> Probably because I copied the code from something that used
> SPLICE_F_NONBLOCK ;-)
>

From the relay API implementation?

> But userspace passes that flag too. Although, maybe I just don't
> understand the API fully.
>

this flag is a little confusing. It is much like:

flags = fcntl(pipe_fd, F_GETFL);
fcntl(pipe_fd, F_SETFL, flags | O_NONBLOCK);
splice(..pipe_fd..., 0); // without O_NONBLOCK
fcntl(pipe_fd, F_SETFL, falgs);

In fact, setting O_NONBLOCK is the only way of letting splice pipe
operations non-blockable currently. When splicing, splice doesn't
refer to the pipe file's f_flags. :(

-- 
Regards，
Changli Gao(xiaosuo@gmail.com)