[PATCH] random: Forbid llseek on random chardev

[PATCH] random: Forbid llseek on random chardev

[Frederic Weisbecker]

From: Arnd Bergmann <arnd@arndb.de>

Seeking on /dev/random and /dev/urandom is pointless.
Using generic_file_llseek means we no longer need to
take the BKL if anyone tries to seek on these.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Matt Mackall <mpm@selenic.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
---
 drivers/char/random.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/drivers/char/random.c b/drivers/char/random.c
index 2fd3d39..513c685 100644
--- a/drivers/char/random.c
+++ b/drivers/char/random.c
@@ -1169,6 +1169,7 @@ const struct file_operations random_fops = {
 	.poll  = random_poll,
 	.unlocked_ioctl = random_ioctl,
 	.fasync = random_fasync,
+	.open = nonseekable_open,
 };
 
 const struct file_operations urandom_fops = {
@@ -1176,6 +1177,7 @@ const struct file_operations urandom_fops = {
 	.write = random_write,
 	.unlocked_ioctl = random_ioctl,
 	.fasync = random_fasync,
+	.open = nonseekable_open,
 };
 
 /***************************************************************
-- 
1.6.2.3

Re: [PATCH] random: Forbid llseek on random chardev

[Matt Mackall]

On Sat, 2010-04-10 at 16:17 +0200, Frederic Weisbecker wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> Seeking on /dev/random and /dev/urandom is pointless.

It is indeed pointless, though that doesn't mean no one does it. 
Forbidding a no-op seems a rather unfriendly way to fix this.

> Using generic_file_llseek means we no longer need to
> take the BKL if anyone tries to seek on these.

Comment doesn't match the patch?

> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> Cc: Matt Mackall <mpm@selenic.com>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
> ---
>  drivers/char/random.c |    2 ++
>  1 files changed, 2 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/char/random.c b/drivers/char/random.c
> index 2fd3d39..513c685 100644
> --- a/drivers/char/random.c
> +++ b/drivers/char/random.c
> @@ -1169,6 +1169,7 @@ const struct file_operations random_fops = {
>  	.poll  = random_poll,
>  	.unlocked_ioctl = random_ioctl,
>  	.fasync = random_fasync,
> +	.open = nonseekable_open,
>  };
>  
>  const struct file_operations urandom_fops = {
> @@ -1176,6 +1177,7 @@ const struct file_operations urandom_fops = {
>  	.write = random_write,
>  	.unlocked_ioctl = random_ioctl,
>  	.fasync = random_fasync,
> +	.open = nonseekable_open,
>  };
>  
>  /***************************************************************



-- 
http://selenic.com : development and support for Mercurial and Linux

Re: [PATCH] random: Forbid llseek on random chardev

[Frederic Weisbecker]

On Sat, Apr 10, 2010 at 11:25:50AM -0500, Matt Mackall wrote:
> On Sat, 2010-04-10 at 16:17 +0200, Frederic Weisbecker wrote:
> > From: Arnd Bergmann <arnd@arndb.de>
> > 
> > Seeking on /dev/random and /dev/urandom is pointless.
> 
> It is indeed pointless, though that doesn't mean no one does it. 
> Forbidding a no-op seems a rather unfriendly way to fix this.


I worried about that too. If you think that could break some user
space apps, we can make it a generic_file_llseek.

 
> > Using generic_file_llseek means we no longer need to
> > take the BKL if anyone tries to seek on these.
> 
> Comment doesn't match the patch?


Oops. I guess the patch has been updated, but not the changelog.

Some background: we are trying to remove the uses of default_llseek
that use the bkl. We started with turning llseek stubs to
generic_file_llseek (or non seekable open in some cases), but we are
hesitating now and think about actually turn all the stubs that
might need the bkl into default_llseek, and fallback to
generic_file_llseek for all other stubs.

Depending on what we do I'll resend you an updated version
of this patch with generic_file_llseek, or I will let the stub
as is.

Thanks.

Re: [PATCH] random: Forbid llseek on random chardev

[Matt Mackall]

On Sat, 2010-04-10 at 18:35 +0200, Frederic Weisbecker wrote:
> On Sat, Apr 10, 2010 at 11:25:50AM -0500, Matt Mackall wrote:
> > On Sat, 2010-04-10 at 16:17 +0200, Frederic Weisbecker wrote:
> > > From: Arnd Bergmann <arnd@arndb.de>
> > > 
> > > Seeking on /dev/random and /dev/urandom is pointless.
> > 
> > It is indeed pointless, though that doesn't mean no one does it. 
> > Forbidding a no-op seems a rather unfriendly way to fix this.
> 
> 
> I worried about that too. If you think that could break some user
> space apps, we can make it a generic_file_llseek.
> 
>  
> > > Using generic_file_llseek means we no longer need to
> > > take the BKL if anyone tries to seek on these.
> > 
> > Comment doesn't match the patch?
> 
> 
> Oops. I guess the patch has been updated, but not the changelog.
> 
> Some background: we are trying to remove the uses of default_llseek
> that use the bkl. We started with turning llseek stubs to
> generic_file_llseek (or non seekable open in some cases), but we are
> hesitating now and think about actually turn all the stubs that
> might need the bkl into default_llseek, and fallback to
> generic_file_llseek for all other stubs.
> 
> Depending on what we do I'll resend you an updated version
> of this patch with generic_file_llseek, or I will let the stub
> as is.

Sounds good.

-- 
http://selenic.com : development and support for Mercurial and Linux

Re: [PATCH] random: Forbid llseek on random chardev

[Arnd Bergmann]

On Saturday 10 April 2010 18:35:30 Frederic Weisbecker wrote:
> 
> > > Using generic_file_llseek means we no longer need to
> > > take the BKL if anyone tries to seek on these.
> > 
> > Comment doesn't match the patch?

Yes, my fault. The version in my series already had this problem.

> Oops. I guess the patch has been updated, but not the changelog.
> 
> Some background: we are trying to remove the uses of default_llseek
> that use the bkl. We started with turning llseek stubs to
> generic_file_llseek (or non seekable open in some cases), but we are
> hesitating now and think about actually turn all the stubs that
> might need the bkl into default_llseek, and fallback to
> generic_file_llseek for all other stubs.
> 
> Depending on what we do I'll resend you an updated version
> of this patch with generic_file_llseek, or I will let the stub
> as is.

The noop_llseek from Jan's series seems most appropriate here.

	Arnd