Hi Andrew, Al,
I posted this patch a few months ago, but it apparently
fell thru cracks. Here we go again.
I noticed that read/write/rdwr_pipe_fops are (1) const and
(2) exactly identical to xxx_fifo_fops, which are also const.
Attached patch #defines xxx_pipe_fops as aliases to xxx_fifo_fops.
Size difference:
# size linux-2.6.25-rc6*/*/pipe.o
text data bss dec hex filename
6534 144 0 6678 1a16 linux-2.6.25-rc6/fs/pipe.o
5862 144 0 6006 1776 linux-2.6.25-rc6-pt/fs/pipe.o
Run-tested on 2.6.26-rc8. Please apply.
Signed-off-by: Denys Vlasenko <[email protected]>
--
vda
On Tue, Jul 01, 2008 at 11:03:02AM +0200, Denys Vlasenko wrote:
> Hi Andrew, Al,
>
> I posted this patch a few months ago, but it apparently
> fell thru cracks. Here we go again.
>
> I noticed that read/write/rdwr_pipe_fops are (1) const and
> (2) exactly identical to xxx_fifo_fops, which are also const.
>
> Attached patch #defines xxx_pipe_fops as aliases to xxx_fifo_fops.
> Size difference:
Why do you need the #defines? Just use the _fifo_ name directly.
On Tue, 1 Jul 2008 11:03:02 +0200 Denys Vlasenko <[email protected]> wrote:
> Hi Andrew, Al,
>
> I posted this patch a few months ago, but it apparently
> fell thru cracks. Here we go again.
>
> I noticed that read/write/rdwr_pipe_fops are (1) const and
> (2) exactly identical to xxx_fifo_fops, which are also const.
>
> Attached patch #defines xxx_pipe_fops as aliases to xxx_fifo_fops.
> Size difference:
>
> # size linux-2.6.25-rc6*/*/pipe.o
> text data bss dec hex filename
> 6534 144 0 6678 1a16 linux-2.6.25-rc6/fs/pipe.o
> 5862 144 0 6006 1776 linux-2.6.25-rc6-pt/fs/pipe.o
>
> Run-tested on 2.6.26-rc8. Please apply.
>
> Signed-off-by: Denys Vlasenko <[email protected]>
<argh, an attachment. save-as, read, edit..>
--- linux-2.6.25-rc6.src/fs/pipe.c Sat Mar 22 23:00:34 2008
> +++ linux-2.6.25-rc6.pipe/fs/pipe.c Fri Mar 28 15:52:00 2008
> @@ -814,42 +814,9 @@
> .fasync = pipe_rdwr_fasync,
> };
>
> -static const struct file_operations read_pipe_fops = {
> - .llseek = no_llseek,
> - .read = do_sync_read,
> - .aio_read = pipe_read,
> - .write = bad_pipe_w,
> - .poll = pipe_poll,
> - .unlocked_ioctl = pipe_ioctl,
> - .open = pipe_read_open,
> - .release = pipe_read_release,
> - .fasync = pipe_read_fasync,
> -};
> -
> -static const struct file_operations write_pipe_fops = {
> - .llseek = no_llseek,
> - .read = bad_pipe_r,
> - .write = do_sync_write,
> - .aio_write = pipe_write,
> - .poll = pipe_poll,
> - .unlocked_ioctl = pipe_ioctl,
> - .open = pipe_write_open,
> - .release = pipe_write_release,
> - .fasync = pipe_write_fasync,
> -};
> -
> -static const struct file_operations rdwr_pipe_fops = {
> - .llseek = no_llseek,
> - .read = do_sync_read,
> - .aio_read = pipe_read,
> - .write = do_sync_write,
> - .aio_write = pipe_write,
> - .poll = pipe_poll,
> - .unlocked_ioctl = pipe_ioctl,
> - .open = pipe_rdwr_open,
> - .release = pipe_rdwr_release,
> - .fasync = pipe_rdwr_fasync,
> -};
> +#define read_pipe_fops read_fifo_fops
> +#define write_pipe_fops write_fifo_fops
> +#define rdwr_pipe_fops rdwr_fifo_fops
>
> struct pipe_inode_info * alloc_pipe_info(struct inode *inode)
> {
Well OK. But there's a risk that someone will go and modify
read_fifo_fops without realising that they're also modifying
read_pipe_fops.
So it'd be better to rename read_fifo_fops to (say) shared_read_fops
then do
#define read_pipe_fops shared_read_fops
#define read_fifo_fops shared_read_fops
no?
On Tuesday 01 July 2008 09:32, Andrew Morton wrote:
> > I noticed that read/write/rdwr_pipe_fops are (1) const and
> > (2) exactly identical to xxx_fifo_fops, which are also const.
> >
> > Attached patch #defines xxx_pipe_fops as aliases to xxx_fifo_fops.
> > Size difference:
> >
> > # size linux-2.6.25-rc6*/*/pipe.o
> > text data bss dec hex filename
> > 6534 144 0 6678 1a16 linux-2.6.25-rc6/fs/pipe.o
> > 5862 144 0 6006 1776 linux-2.6.25-rc6-pt/fs/pipe.o
> >
> > Run-tested on 2.6.26-rc8. Please apply.
> >
> > Signed-off-by: Denys Vlasenko <[email protected]>
>
> <argh, an attachment. save-as, read, edit..>
Let's see how KMail will cope with inline cut-n-paste...
> > -static const struct file_operations rdwr_pipe_fops = {
> > - .llseek = no_llseek,
> > - .read = do_sync_read,
> > - .aio_read = pipe_read,
> > - .write = do_sync_write,
> > - .aio_write = pipe_write,
> > - .poll = pipe_poll,
> > - .unlocked_ioctl = pipe_ioctl,
> > - .open = pipe_rdwr_open,
> > - .release = pipe_rdwr_release,
> > - .fasync = pipe_rdwr_fasync,
> > -};
> > +#define read_pipe_fops read_fifo_fops
> > +#define write_pipe_fops write_fifo_fops
> > +#define rdwr_pipe_fops rdwr_fifo_fops
>
> Well OK. But there's a risk that someone will go and modify
> read_fifo_fops without realising that they're also modifying
> read_pipe_fops.
Yes, it is possible.
> So it'd be better to rename read_fifo_fops to (say) shared_read_fops
> then do
>
> #define read_pipe_fops shared_read_fops
> #define read_fifo_fops shared_read_fops
I think since XXX_pipe_fops are only used in this file,
just explaining this in the comment would be enough.
Please take a look at the version below.
Signed-off-by: Denys Vlasenko <[email protected]>
--
vda
--- linux-2.6.26-rc8/fs/pipe.c.org Tue Jul 1 11:52:28 2008
+++ linux-2.6.26-rc8/fs/pipe.c Tue Jul 1 11:58:23 2008
@@ -777,6 +777,8 @@
/*
* The file_operations structs are not static because they
* are also used in linux/fs/fifo.c to do operations on FIFOs.
+ *
+ * Pipes reuse fifos' file_operations structs.
*/
const struct file_operations read_fifo_fops = {
.llseek = no_llseek,
@@ -815,43 +817,6 @@
.fasync = pipe_rdwr_fasync,
};
-static const struct file_operations read_pipe_fops = {
- .llseek = no_llseek,
- .read = do_sync_read,
- .aio_read = pipe_read,
- .write = bad_pipe_w,
- .poll = pipe_poll,
- .unlocked_ioctl = pipe_ioctl,
- .open = pipe_read_open,
- .release = pipe_read_release,
- .fasync = pipe_read_fasync,
-};
-
-static const struct file_operations write_pipe_fops = {
- .llseek = no_llseek,
- .read = bad_pipe_r,
- .write = do_sync_write,
- .aio_write = pipe_write,
- .poll = pipe_poll,
- .unlocked_ioctl = pipe_ioctl,
- .open = pipe_write_open,
- .release = pipe_write_release,
- .fasync = pipe_write_fasync,
-};
-
-static const struct file_operations rdwr_pipe_fops = {
- .llseek = no_llseek,
- .read = do_sync_read,
- .aio_read = pipe_read,
- .write = do_sync_write,
- .aio_write = pipe_write,
- .poll = pipe_poll,
- .unlocked_ioctl = pipe_ioctl,
- .open = pipe_rdwr_open,
- .release = pipe_rdwr_release,
- .fasync = pipe_rdwr_fasync,
-};
-
struct pipe_inode_info * alloc_pipe_info(struct inode *inode)
{
struct pipe_inode_info *pipe;
@@ -927,7 +892,7 @@
inode->i_pipe = pipe;
pipe->readers = pipe->writers = 1;
- inode->i_fop = &rdwr_pipe_fops;
+ inode->i_fop = &rdwr_fifo_fops;
/*
* Mark the inode dirty from the very beginning,
@@ -978,7 +943,7 @@
d_instantiate(dentry, inode);
err = -ENFILE;
- f = alloc_file(pipe_mnt, dentry, FMODE_WRITE, &write_pipe_fops);
+ f = alloc_file(pipe_mnt, dentry, FMODE_WRITE, &write_fifo_fops);
if (!f)
goto err_dentry;
f->f_mapping = inode->i_mapping;
@@ -1021,7 +986,7 @@
f->f_pos = 0;
f->f_flags = O_RDONLY;
- f->f_op = &read_pipe_fops;
+ f->f_op = &read_fifo_fops;
f->f_mode = FMODE_READ;
f->f_version = 0;
On Tue, 1 Jul 2008 12:03:50 +0200 Denys Vlasenko <[email protected]> wrote:
> I think since XXX_pipe_fops are only used in this file,
> just explaining this in the comment would be enough.
no, a comment is only needed when the code is unobvious. Make
the code obvious and we don't need a comment.
As Christoph pointed out, open-coding shared_read_fops everywhere
might make sense too. It'd make it harder to unshare them later
on, but that's pretty improbable.
On Tue, Jul 01, 2008 at 01:10:45AM -0700, Andrew Morton wrote:
> On Tue, 1 Jul 2008 12:03:50 +0200 Denys Vlasenko <[email protected]> wrote:
>
> > I think since XXX_pipe_fops are only used in this file,
> > just explaining this in the comment would be enough.
>
> no, a comment is only needed when the code is unobvious. Make
> the code obvious and we don't need a comment.
>
> As Christoph pointed out, open-coding shared_read_fops everywhere
> might make sense too. It'd make it harder to unshare them later
> on, but that's pretty improbable.
Deny's latest patch doesn't have the #defines anymore and looks quite
reasonable to me. If you really insist on magic naming I think shared
is a very bad choice because it doesn't have any useful meaning in this
context. If you insist on magic naming do _pipefifo_, but I think
the current version with the comment is much better.
On Tuesday 01 July 2008 10:10, Andrew Morton wrote:
> On Tue, 1 Jul 2008 12:03:50 +0200 Denys Vlasenko <[email protected]> wrote:
>
> > I think since XXX_pipe_fops are only used in this file,
> > just explaining this in the comment would be enough.
>
> no, a comment is only needed when the code is unobvious. Make
> the code obvious and we don't need a comment.
>
> As Christoph pointed out, open-coding shared_read_fops everywhere
> might make sense too. It'd make it harder to unshare them later
> on, but that's pretty improbable.
Ok.
Signed-off-by: Denys Vlasenko <[email protected]>
--
vda
--- linux-2.6.26-rc8/fs.org/fifo.c Thu Apr 17 04:49:44 2008
+++ linux-2.6.26-rc8/fs/fifo.c Tue Jul 1 14:09:13 2008
@@ -57,7 +57,7 @@
* POSIX.1 says that O_NONBLOCK means return with the FIFO
* opened, even when there is no process writing the FIFO.
*/
- filp->f_op = &read_fifo_fops;
+ filp->f_op = &read_pipefifo_fops;
pipe->r_counter++;
if (pipe->readers++ == 0)
wake_up_partner(inode);
@@ -86,7 +86,7 @@
if ((filp->f_flags & O_NONBLOCK) && !pipe->readers)
goto err;
- filp->f_op = &write_fifo_fops;
+ filp->f_op = &write_pipefifo_fops;
pipe->w_counter++;
if (!pipe->writers++)
wake_up_partner(inode);
@@ -105,7 +105,7 @@
* This implementation will NEVER block on a O_RDWR open, since
* the process can at least talk to itself.
*/
- filp->f_op = &rdwr_fifo_fops;
+ filp->f_op = &rdwr_pipefifo_fops;
pipe->readers++;
pipe->writers++;
@@ -151,5 +151,5 @@
* depending on the access mode of the file...
*/
const struct file_operations def_fifo_fops = {
- .open = fifo_open, /* will set read or write pipe_fops */
+ .open = fifo_open, /* will set read_ or write_pipefifo_fops */
};
--- linux-2.6.26-rc8/fs.org/pipe.c Tue Jul 1 11:52:28 2008
+++ linux-2.6.26-rc8/fs/pipe.c Tue Jul 1 14:08:16 2008
@@ -777,8 +777,10 @@
/*
* The file_operations structs are not static because they
* are also used in linux/fs/fifo.c to do operations on FIFOs.
+ *
+ * Pipes reuse fifos' file_operations structs.
*/
-const struct file_operations read_fifo_fops = {
+const struct file_operations read_pipefifo_fops = {
.llseek = no_llseek,
.read = do_sync_read,
.aio_read = pipe_read,
@@ -790,7 +792,7 @@
.fasync = pipe_read_fasync,
};
-const struct file_operations write_fifo_fops = {
+const struct file_operations write_pipefifo_fops = {
.llseek = no_llseek,
.read = bad_pipe_r,
.write = do_sync_write,
@@ -802,7 +804,7 @@
.fasync = pipe_write_fasync,
};
-const struct file_operations rdwr_fifo_fops = {
+const struct file_operations rdwr_pipefifo_fops = {
.llseek = no_llseek,
.read = do_sync_read,
.aio_read = pipe_read,
@@ -815,43 +817,6 @@
.fasync = pipe_rdwr_fasync,
};
-static const struct file_operations read_pipe_fops = {
- .llseek = no_llseek,
- .read = do_sync_read,
- .aio_read = pipe_read,
- .write = bad_pipe_w,
- .poll = pipe_poll,
- .unlocked_ioctl = pipe_ioctl,
- .open = pipe_read_open,
- .release = pipe_read_release,
- .fasync = pipe_read_fasync,
-};
-
-static const struct file_operations write_pipe_fops = {
- .llseek = no_llseek,
- .read = bad_pipe_r,
- .write = do_sync_write,
- .aio_write = pipe_write,
- .poll = pipe_poll,
- .unlocked_ioctl = pipe_ioctl,
- .open = pipe_write_open,
- .release = pipe_write_release,
- .fasync = pipe_write_fasync,
-};
-
-static const struct file_operations rdwr_pipe_fops = {
- .llseek = no_llseek,
- .read = do_sync_read,
- .aio_read = pipe_read,
- .write = do_sync_write,
- .aio_write = pipe_write,
- .poll = pipe_poll,
- .unlocked_ioctl = pipe_ioctl,
- .open = pipe_rdwr_open,
- .release = pipe_rdwr_release,
- .fasync = pipe_rdwr_fasync,
-};
-
struct pipe_inode_info * alloc_pipe_info(struct inode *inode)
{
struct pipe_inode_info *pipe;
@@ -927,7 +892,7 @@
inode->i_pipe = pipe;
pipe->readers = pipe->writers = 1;
- inode->i_fop = &rdwr_pipe_fops;
+ inode->i_fop = &rdwr_pipefifo_fops;
/*
* Mark the inode dirty from the very beginning,
@@ -978,7 +943,7 @@
d_instantiate(dentry, inode);
err = -ENFILE;
- f = alloc_file(pipe_mnt, dentry, FMODE_WRITE, &write_pipe_fops);
+ f = alloc_file(pipe_mnt, dentry, FMODE_WRITE, &write_pipefifo_fops);
if (!f)
goto err_dentry;
f->f_mapping = inode->i_mapping;
@@ -1021,7 +986,7 @@
f->f_pos = 0;
f->f_flags = O_RDONLY;
- f->f_op = &read_pipe_fops;
+ f->f_op = &read_pipefifo_fops;
f->f_mode = FMODE_READ;
f->f_version = 0;
On Tuesday 01 July 2008 14:11, Denys Vlasenko wrote:
> On Tuesday 01 July 2008 10:10, Andrew Morton wrote:
> > On Tue, 1 Jul 2008 12:03:50 +0200 Denys Vlasenko <[email protected]> wrote:
> >
> > > I think since XXX_pipe_fops are only used in this file,
> > > just explaining this in the comment would be enough.
> >
> > no, a comment is only needed when the code is unobvious. Make
> > the code obvious and we don't need a comment.
> >
> > As Christoph pointed out, open-coding shared_read_fops everywhere
> > might make sense too. It'd make it harder to unshare them later
> > on, but that's pretty improbable.
>
> Ok.
And this one is even better [because it actually compiles :) ].
I forgot to change fs.h - XXX_pipefifo_fops live there.
Signed-off-by: Denys Vlasenko <[email protected]>
--
vda
--- linux-2.6.26-rc8/fs.org/fifo.c Thu Apr 17 04:49:44 2008
+++ linux-2.6.26-rc8/fs/fifo.c Tue Jul 1 14:09:13 2008
@@ -57,7 +57,7 @@
* POSIX.1 says that O_NONBLOCK means return with the FIFO
* opened, even when there is no process writing the FIFO.
*/
- filp->f_op = &read_fifo_fops;
+ filp->f_op = &read_pipefifo_fops;
pipe->r_counter++;
if (pipe->readers++ == 0)
wake_up_partner(inode);
@@ -86,7 +86,7 @@
if ((filp->f_flags & O_NONBLOCK) && !pipe->readers)
goto err;
- filp->f_op = &write_fifo_fops;
+ filp->f_op = &write_pipefifo_fops;
pipe->w_counter++;
if (!pipe->writers++)
wake_up_partner(inode);
@@ -105,7 +105,7 @@
* This implementation will NEVER block on a O_RDWR open, since
* the process can at least talk to itself.
*/
- filp->f_op = &rdwr_fifo_fops;
+ filp->f_op = &rdwr_pipefifo_fops;
pipe->readers++;
pipe->writers++;
@@ -151,5 +151,5 @@
* depending on the access mode of the file...
*/
const struct file_operations def_fifo_fops = {
- .open = fifo_open, /* will set read or write pipe_fops */
+ .open = fifo_open, /* will set read_ or write_pipefifo_fops */
};
--- linux-2.6.26-rc8/fs.org/pipe.c Tue Jul 1 11:52:28 2008
+++ linux-2.6.26-rc8/fs/pipe.c Tue Jul 1 14:08:16 2008
@@ -777,8 +777,10 @@
/*
* The file_operations structs are not static because they
* are also used in linux/fs/fifo.c to do operations on FIFOs.
+ *
+ * Pipes reuse fifos' file_operations structs.
*/
-const struct file_operations read_fifo_fops = {
+const struct file_operations read_pipefifo_fops = {
.llseek = no_llseek,
.read = do_sync_read,
.aio_read = pipe_read,
@@ -790,7 +792,7 @@
.fasync = pipe_read_fasync,
};
-const struct file_operations write_fifo_fops = {
+const struct file_operations write_pipefifo_fops = {
.llseek = no_llseek,
.read = bad_pipe_r,
.write = do_sync_write,
@@ -802,7 +804,7 @@
.fasync = pipe_write_fasync,
};
-const struct file_operations rdwr_fifo_fops = {
+const struct file_operations rdwr_pipefifo_fops = {
.llseek = no_llseek,
.read = do_sync_read,
.aio_read = pipe_read,
@@ -815,43 +817,6 @@
.fasync = pipe_rdwr_fasync,
};
-static const struct file_operations read_pipe_fops = {
- .llseek = no_llseek,
- .read = do_sync_read,
- .aio_read = pipe_read,
- .write = bad_pipe_w,
- .poll = pipe_poll,
- .unlocked_ioctl = pipe_ioctl,
- .open = pipe_read_open,
- .release = pipe_read_release,
- .fasync = pipe_read_fasync,
-};
-
-static const struct file_operations write_pipe_fops = {
- .llseek = no_llseek,
- .read = bad_pipe_r,
- .write = do_sync_write,
- .aio_write = pipe_write,
- .poll = pipe_poll,
- .unlocked_ioctl = pipe_ioctl,
- .open = pipe_write_open,
- .release = pipe_write_release,
- .fasync = pipe_write_fasync,
-};
-
-static const struct file_operations rdwr_pipe_fops = {
- .llseek = no_llseek,
- .read = do_sync_read,
- .aio_read = pipe_read,
- .write = do_sync_write,
- .aio_write = pipe_write,
- .poll = pipe_poll,
- .unlocked_ioctl = pipe_ioctl,
- .open = pipe_rdwr_open,
- .release = pipe_rdwr_release,
- .fasync = pipe_rdwr_fasync,
-};
-
struct pipe_inode_info * alloc_pipe_info(struct inode *inode)
{
struct pipe_inode_info *pipe;
@@ -927,7 +892,7 @@
inode->i_pipe = pipe;
pipe->readers = pipe->writers = 1;
- inode->i_fop = &rdwr_pipe_fops;
+ inode->i_fop = &rdwr_pipefifo_fops;
/*
* Mark the inode dirty from the very beginning,
@@ -978,7 +943,7 @@
d_instantiate(dentry, inode);
err = -ENFILE;
- f = alloc_file(pipe_mnt, dentry, FMODE_WRITE, &write_pipe_fops);
+ f = alloc_file(pipe_mnt, dentry, FMODE_WRITE, &write_pipefifo_fops);
if (!f)
goto err_dentry;
f->f_mapping = inode->i_mapping;
@@ -1021,7 +986,7 @@
f->f_pos = 0;
f->f_flags = O_RDONLY;
- f->f_op = &read_pipe_fops;
+ f->f_op = &read_pipefifo_fops;
f->f_mode = FMODE_READ;
f->f_version = 0;
--- linux-2.6.26-rc8/include.org/linux/fs.h Mon Jun 30 15:45:56 2008
+++ linux-2.6.26-rc8/include/linux/fs.h Tue Jul 1 14:13:22 2008
@@ -1687,9 +1687,9 @@
extern void make_bad_inode(struct inode *);
extern int is_bad_inode(struct inode *);
-extern const struct file_operations read_fifo_fops;
-extern const struct file_operations write_fifo_fops;
-extern const struct file_operations rdwr_fifo_fops;
+extern const struct file_operations read_pipefifo_fops;
+extern const struct file_operations write_pipefifo_fops;
+extern const struct file_operations rdwr_pipefifo_fops;
extern int fs_may_remount_ro(struct super_block *);