Re: + mm-search_binary_handler-mem-limit-fix.patch added to -mm tree

Re: + mm-search_binary_handler-mem-limit-fix.patch added to -mm tree

[Heiko Carstens]

On Wed, Jan 24, 2007 at 01:09:18AM -0800, akpm@osdl.org wrote:
> 
> The patch titled
>      mm: search_binary_handler() mem limit fix
> has been added to the -mm tree.  Its filename is
>      mm-search_binary_handler-mem-limit-fix.patch
> 
> *** Remember to use Documentation/SubmitChecklist when testing your code ***
> 
> See http://www.zip.com.au/~akpm/linux/patches/stuff/added-to-mm.txt to find
> out what to do about this
> 
> ------------------------------------------------------
> Subject: mm: search_binary_handler() mem limit fix
> From: Dmitriy Monakhov <dmonakhov@openvz.org>
> 
> The function changes mem limit to USER_DS before possible modprobe, but
> never restored it again.
> 
> Signed-off-by: Dmitriy Monakhov <dmonakhov@openvz.org>
> Signed-off-by: Andrew Morton <akpm@osdl.org>

This patch breaks s390. I haven't yet tried to figure out why, but does this
patch actually fix a real bug?
What happens is that the init process gets killed -> panic.

>  fs/exec.c |   10 +++++++---
>  1 file changed, 7 insertions(+), 3 deletions(-)
> 
> diff -puN fs/exec.c~mm-search_binary_handler-mem-limit-fix fs/exec.c
> --- a/fs/exec.c~mm-search_binary_handler-mem-limit-fix
> +++ a/fs/exec.c
> @@ -1018,6 +1018,7 @@ int search_binary_handler(struct linux_b
>  {
>  	int try,retval;
>  	struct linux_binfmt *fmt;
> +	mm_segment_t oldfs;
>  #ifdef __alpha__
>  	/* handle /sbin/loader.. */
>  	{
> @@ -1059,11 +1060,12 @@ int search_binary_handler(struct linux_b
> 
>  	/* kernel module loader fixup */
>  	/* so we don't try to load run modprobe in kernel space. */
> +	oldfs = get_fs();
>  	set_fs(USER_DS);
> 
>  	retval = audit_bprm(bprm);
>  	if (retval)
> -		return retval;
> +		goto out;
> 
>  	retval = -ENOENT;
>  	for (try=0; try<2; try++) {
> @@ -1084,7 +1086,7 @@ int search_binary_handler(struct linux_b
>  				bprm->file = NULL;
>  				current->did_exec = 1;
>  				proc_exec_connector(current);
> -				return retval;
> +				goto out;
>  			}
>  			read_lock(&binfmt_lock);
>  			put_binfmt(fmt);
> @@ -1092,7 +1094,7 @@ int search_binary_handler(struct linux_b
>  				break;
>  			if (!bprm->file) {
>  				read_unlock(&binfmt_lock);
> -				return retval;
> +				goto out;
>  			}
>  		}
>  		read_unlock(&binfmt_lock);
> @@ -1110,6 +1112,8 @@ int search_binary_handler(struct linux_b
>  #endif
>  		}
>  	}
> +out:
> +	set_fs(oldfs);
>  	return retval;
>  }

Re: + mm-search_binary_handler-mem-limit-fix.patch added to -mm tree

[Heiko Carstens]

On Mon, Jan 29, 2007 at 12:33:28PM +0100, Heiko Carstens wrote:
> On Wed, Jan 24, 2007 at 01:09:18AM -0800, akpm@osdl.org wrote:
> >      mm-search_binary_handler-mem-limit-fix.patch
> > From: Dmitriy Monakhov <dmonakhov@openvz.org>
> > 
> > The function changes mem limit to USER_DS before possible modprobe, but
> > never restored it again.
> > 
> > Signed-off-by: Dmitriy Monakhov <dmonakhov@openvz.org>
> > Signed-off-by: Andrew Morton <akpm@osdl.org>
> >  fs/exec.c |   10 +++++++---
> >  1 file changed, 7 insertions(+), 3 deletions(-)
> > 
> > diff -puN fs/exec.c~mm-search_binary_handler-mem-limit-fix fs/exec.c
> > --- a/fs/exec.c~mm-search_binary_handler-mem-limit-fix
> > +++ a/fs/exec.c
> > @@ -1018,6 +1018,7 @@ int search_binary_handler(struct linux_b
> >  {
> >  	int try,retval;
> >  	struct linux_binfmt *fmt;
> > +	mm_segment_t oldfs;
> >  #ifdef __alpha__
> >  	/* handle /sbin/loader.. */
> >  	{
> > @@ -1059,11 +1060,12 @@ int search_binary_handler(struct linux_b
> > 
> >  	/* kernel module loader fixup */
> >  	/* so we don't try to load run modprobe in kernel space. */
> > +	oldfs = get_fs();
> >  	set_fs(USER_DS);
> > 
> >  	retval = audit_bprm(bprm);
> >  	if (retval)
> > -		return retval;
> > +		goto out;
> > 
> >  	retval = -ENOENT;
> >  	for (try=0; try<2; try++) {
> > @@ -1084,7 +1086,7 @@ int search_binary_handler(struct linux_b
> >  				bprm->file = NULL;
> >  				current->did_exec = 1;
> >  				proc_exec_connector(current);
> > -				return retval;
> > +				goto out;
> >  			}
> >  			read_lock(&binfmt_lock);
> >  			put_binfmt(fmt);
> > @@ -1092,7 +1094,7 @@ int search_binary_handler(struct linux_b
> >  				break;
> >  			if (!bprm->file) {
> >  				read_unlock(&binfmt_lock);
> > -				return retval;
> > +				goto out;
> >  			}
> >  		}
> >  		read_unlock(&binfmt_lock);
> > @@ -1110,6 +1112,8 @@ int search_binary_handler(struct linux_b
> >  #endif
> >  		}
> >  	}
> > +out:
> > +	set_fs(oldfs);
> >  	return retval;
> >  }

This is broken. This is the only place in kernel that sets fs to USER_DS
for a new process. With this patch we could as well get rid of USER_DS and
all the address space checkings.
Besides that it breaks architectures with distinct physical address spaces.

Re: + mm-search_binary_handler-mem-limit-fix.patch added to -mm tree

[Andrew Morton]

On Mon, 29 Jan 2007 14:59:58 +0100
Heiko Carstens <heiko.carstens@de.ibm.com> wrote:

> On Mon, Jan 29, 2007 at 12:33:28PM +0100, Heiko Carstens wrote:
> > On Wed, Jan 24, 2007 at 01:09:18AM -0800, akpm@osdl.org wrote:
> > >      mm-search_binary_handler-mem-limit-fix.patch
> > > From: Dmitriy Monakhov <dmonakhov@openvz.org>
> > > 
> > > The function changes mem limit to USER_DS before possible modprobe, but
> > > never restored it again.
> > > 
> > > Signed-off-by: Dmitriy Monakhov <dmonakhov@openvz.org>
> > > Signed-off-by: Andrew Morton <akpm@osdl.org>
> > >  fs/exec.c |   10 +++++++---
> > >  1 file changed, 7 insertions(+), 3 deletions(-)
> > > 
> > > diff -puN fs/exec.c~mm-search_binary_handler-mem-limit-fix fs/exec.c
> > > --- a/fs/exec.c~mm-search_binary_handler-mem-limit-fix
> > > +++ a/fs/exec.c
> > > @@ -1018,6 +1018,7 @@ int search_binary_handler(struct linux_b
> > >  {
> > >  	int try,retval;
> > >  	struct linux_binfmt *fmt;
> > > +	mm_segment_t oldfs;
> > >  #ifdef __alpha__
> > >  	/* handle /sbin/loader.. */
> > >  	{
> > > @@ -1059,11 +1060,12 @@ int search_binary_handler(struct linux_b
> > > 
> > >  	/* kernel module loader fixup */
> > >  	/* so we don't try to load run modprobe in kernel space. */
> > > +	oldfs = get_fs();
> > >  	set_fs(USER_DS);
> > > 
> > >  	retval = audit_bprm(bprm);
> > >  	if (retval)
> > > -		return retval;
> > > +		goto out;
> > > 
> > >  	retval = -ENOENT;
> > >  	for (try=0; try<2; try++) {
> > > @@ -1084,7 +1086,7 @@ int search_binary_handler(struct linux_b
> > >  				bprm->file = NULL;
> > >  				current->did_exec = 1;
> > >  				proc_exec_connector(current);
> > > -				return retval;
> > > +				goto out;
> > >  			}
> > >  			read_lock(&binfmt_lock);
> > >  			put_binfmt(fmt);
> > > @@ -1092,7 +1094,7 @@ int search_binary_handler(struct linux_b
> > >  				break;
> > >  			if (!bprm->file) {
> > >  				read_unlock(&binfmt_lock);
> > > -				return retval;
> > > +				goto out;
> > >  			}
> > >  		}
> > >  		read_unlock(&binfmt_lock);
> > > @@ -1110,6 +1112,8 @@ int search_binary_handler(struct linux_b
> > >  #endif
> > >  		}
> > >  	}
> > > +out:
> > > +	set_fs(oldfs);
> > >  	return retval;
> > >  }
> 
> This is broken. This is the only place in kernel that sets fs to USER_DS
> for a new process. With this patch we could as well get rid of USER_DS and
> all the address space checkings.
> Besides that it breaks architectures with distinct physical address spaces.

hm, thanks for testing - I'll drop it.

I don't really understand what's wrong with it though.  Maybe it's settng
USER_DS on kernel threads?

Re: + mm-search_binary_handler-mem-limit-fix.patch added to -mm tree

[Martin Schwidefsky]

On Mon, 2007-01-29 at 09:37 -0800, Andrew Morton wrote:
> hm, thanks for testing - I'll drop it.
> 
> I don't really understand what's wrong with it though.  Maybe it's settng
> USER_DS on kernel threads?

For architectures with a split address space there has to be a call
set_fs(USER_DS) that switches from KERNEL_DS to USER_DS for the init
process. So far this has been done in search_binary_handler and
traditionally the kernel starts with KERNEL_DS to make the early
copy_from_user calls work.
So, what is wrong with always setting USER_DS? We are starting a user
space process after all.

-- 
blue skies,
  Martin.

Martin Schwidefsky
Linux for zSeries Development & Services
IBM Deutschland Entwicklung GmbH

"Reality continues to ruin my life." - Calvin.

Re: + mm-search_binary_handler-mem-limit-fix.patch added to -mm tree

[Dmitriy Monakhov]

> > > The function changes mem limit to USER_DS before possible modprobe, but
> > > never restored it again.
Truly. The road to hell is paved with good intentions.

Martin Schwidefsky <schwidefsky@de.ibm.com> writes:
> On Mon, 2007-01-29 at 09:37 -0800, Andrew Morton wrote:
>> hm, thanks for testing - I'll drop it.
>> 
>> I don't really understand what's wrong with it though.  Maybe it's settng
>> USER_DS on kernel threads?
>
> For architectures with a split address space there has to be a call
> set_fs(USER_DS) that switches from KERNEL_DS to USER_DS for the init
> process. So far this has been done in search_binary_handler and
> traditionally the kernel starts with KERNEL_DS to make the early
> copy_from_user calls work.
> So, what is wrong with always setting USER_DS? We are starting a user
> space process after all.
May be add some comment to prevent future attempts to make this place 
more "correct"?
>
> -- 
> blue skies,
>   Martin.
>
> Martin Schwidefsky
> Linux for zSeries Development & Services
> IBM Deutschland Entwicklung GmbH
>
> "Reality continues to ruin my life." - Calvin.
>
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

Re: + mm-search_binary_handler-mem-limit-fix.patch added to -mm tree

[Martin Schwidefsky]

On Tue, 2007-01-30 at 08:40 +0300, Dmitriy Monakhov wrote:
> > > > The function changes mem limit to USER_DS before possible modprobe, but
> > > > never restored it again.
> Truly. The road to hell is paved with good intentions.

:-)

> > For architectures with a split address space there has to be a call
> > set_fs(USER_DS) that switches from KERNEL_DS to USER_DS for the init
> > process. So far this has been done in search_binary_handler and
> > traditionally the kernel starts with KERNEL_DS to make the early
> > copy_from_user calls work.
> > So, what is wrong with always setting USER_DS? We are starting a user
> > space process after all.
> May be add some comment to prevent future attempts to make this place 
> more "correct"?

The use of set_fs(USER_DS) in search_binary_handler is certainly
different compared to the rest. It probably is the only one that is not
paired with a set_fs(KERNEL_DS) or set_fs(old_fs). A comment won't hurt.

-- 
blue skies,
  Martin.

Martin Schwidefsky
Linux for zSeries Development & Services
IBM Deutschland Entwicklung GmbH

"Reality continues to ruin my life." - Calvin.