[PATCH 1/1] coredump: use task comm instead of (unknown)

[PATCH 1/1] coredump: use task comm instead of (unknown)

[Jiri Slaby]

If we don't know the file corresponding to the binary (i.e. exe_file
is unknown), use "task->comm (path unknown)" instead of simple
"(unknown)" as suggested by ak.

The fallback is the same as %e except it will append "(path unknown)".

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andi Kleen <andi@firstfloor.org>
---
 fs/exec.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/fs/exec.c b/fs/exec.c
index 5ee7562..0a4d281 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -1555,7 +1555,7 @@ static int cn_print_exe_file(struct core_name *cn)
 
 	exe_file = get_mm_exe_file(current->mm);
 	if (!exe_file)
-		return cn_printf(cn, "(unknown)");
+		return cn_printf(cn, "%s (path unknown)", current->comm);
 
 	pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
 	if (!pathbuf) {
-- 
1.7.4.2

Re: [PATCH 1/1] coredump: use task comm instead of (unknown)

[Andrew Morton]

On Wed,  4 May 2011 09:32:34 +0200
Jiri Slaby <jslaby@suse.cz> wrote:

> If we don't know the file corresponding to the binary (i.e. exe_file
> is unknown), use "task->comm (path unknown)" instead of simple
> "(unknown)" as suggested by ak.
> 
> The fallback is the same as %e except it will append "(path unknown)".
> 
> Signed-off-by: Jiri Slaby <jslaby@suse.cz>
> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
> Cc: Al Viro <viro@zeniv.linux.org.uk>
> Cc: Andi Kleen <andi@firstfloor.org>
> ---
>  fs/exec.c |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)
> 
> diff --git a/fs/exec.c b/fs/exec.c
> index 5ee7562..0a4d281 100644
> --- a/fs/exec.c
> +++ b/fs/exec.c
> @@ -1555,7 +1555,7 @@ static int cn_print_exe_file(struct core_name *cn)
>  
>  	exe_file = get_mm_exe_file(current->mm);
>  	if (!exe_file)
> -		return cn_printf(cn, "(unknown)");
> +		return cn_printf(cn, "%s (path unknown)", current->comm);
>  
>  	pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
>  	if (!pathbuf) {

Direct access to current->comm is racy since we added
prctl(PR_SET_NAME).

Hopefully John Stultz will soon be presenting us with a %p modifier for
displaying task_struct.comm.

But we should get this settled pretty promptly as this is a form of
userspace-visible API.  Use get_task_comm() for now.

Also, there's nothing which prevents userspace from rewriting
task->comm to something which contains slashes (this seems bad).  If
that is done, your patch will do Bad Things - it should be modified to
use cn_print_exe_file()'s slash-overwriting codepath.

Re: [PATCH 1/1] coredump: use task comm instead of (unknown)

[Jiri Slaby]

Ccing Oleg.

On 05/06/2011 12:06 AM, Andrew Morton wrote:
> On Wed,  4 May 2011 09:32:34 +0200
> Jiri Slaby <jslaby@suse.cz> wrote:
> 
>> If we don't know the file corresponding to the binary (i.e. exe_file
>> is unknown), use "task->comm (path unknown)" instead of simple
>> "(unknown)" as suggested by ak.
>>
>> The fallback is the same as %e except it will append "(path unknown)".
>>
>> Signed-off-by: Jiri Slaby <jslaby@suse.cz>
>> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
>> Cc: Al Viro <viro@zeniv.linux.org.uk>
>> Cc: Andi Kleen <andi@firstfloor.org>
>> ---
>>  fs/exec.c |    2 +-
>>  1 files changed, 1 insertions(+), 1 deletions(-)
>>
>> diff --git a/fs/exec.c b/fs/exec.c
>> index 5ee7562..0a4d281 100644
>> --- a/fs/exec.c
>> +++ b/fs/exec.c
>> @@ -1555,7 +1555,7 @@ static int cn_print_exe_file(struct core_name *cn)
>>  
>>  	exe_file = get_mm_exe_file(current->mm);
>>  	if (!exe_file)
>> -		return cn_printf(cn, "(unknown)");
>> +		return cn_printf(cn, "%s (path unknown)", current->comm);
>>  
>>  	pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
>>  	if (!pathbuf) {
> 
> Direct access to current->comm is racy since we added
> prctl(PR_SET_NAME).
> 
> Hopefully John Stultz will soon be presenting us with a %p modifier for
> displaying task_struct.comm.

Then just make sure, you won't nest alloc_lock (task_lock) into siglock.

> But we should get this settled pretty promptly as this is a form of
> userspace-visible API.  Use get_task_comm() for now.

I thought about using get_task_comm, but was not sure, if it is safe to
task_lock() at that place. Note that this is copied from %e.

> Also, there's nothing which prevents userspace from rewriting
> task->comm to something which contains slashes (this seems bad).  If
> that is done, your patch will do Bad Things - it should be modified to
> use cn_print_exe_file()'s slash-overwriting codepath.

%E (cn_print_exe_file) does exactly what %e (format_corename) does. So
if this is really broken in the two ways, we should fix both the old %e
and the new %E.

I'm not sure whether at this point when the task is being killed and
dumped, it can still change comm?

For the slashes, I agree. That should be fixed in both cases.

thanks,
-- 
js
suse labs

Re: [PATCH 1/1] coredump: use task comm instead of (unknown)

[Tetsuo Handa]

Andrew Morton wrote:
> Direct access to current->comm is racy since we added
> prctl(PR_SET_NAME).

A passer-by's question. Did we change below one

char *get_task_comm(char *buf, struct task_struct *tsk)
{
	/* buf must be at least sizeof(tsk->comm) in size */
	task_lock(tsk);
	strncpy(buf, tsk->comm, sizeof(tsk->comm));
	task_unlock(tsk);
	return buf;
}

since some archs may do like

char *strncpy(char *dest, const char *src, size_t n)
{
	size_t len = __strnend(src, n) - src;
	__builtin_memset(dest + len, 0, n - len);
	__builtin_memcpy(dest, src, len);
	return dest;
}

( arch/s390/lib/string.c ) after adding prctl(PR_SET_NAME)?

Re: [PATCH 1/1] coredump: use task comm instead of (unknown)

[Andrew Morton]

(Cc John)

On Fri, 6 May 2011 21:26:39 +0900
Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> wrote:

> Andrew Morton wrote:
> > Direct access to current->comm is racy since we added
> > prctl(PR_SET_NAME).
> 
> A passer-by's question. Did we change below one
> 
> char *get_task_comm(char *buf, struct task_struct *tsk)
> {
> 	/* buf must be at least sizeof(tsk->comm) in size */
> 	task_lock(tsk);
> 	strncpy(buf, tsk->comm, sizeof(tsk->comm));
> 	task_unlock(tsk);
> 	return buf;
> }
> 
> since some archs may do like
> 
> char *strncpy(char *dest, const char *src, size_t n)
> {
> 	size_t len = __strnend(src, n) - src;
> 	__builtin_memset(dest + len, 0, n - len);
> 	__builtin_memcpy(dest, src, len);
> 	return dest;
> }
> 
> ( arch/s390/lib/string.c ) after adding prctl(PR_SET_NAME)?

I don't understand what you're asking, sorry.

Re: [PATCH 1/1] coredump: use task comm instead of (unknown)

[Tetsuo Handa]

Andrew Morton wrote:
> > char *strncpy(char *dest, const char *src, size_t n)
> > {
> > 	size_t len = __strnend(src, n) - src;

If src was overwritten by prctl(PR_SET_NAME) at this moment (i.e. after len was
calculated),

> > 	__builtin_memset(dest + len, 0, n - len);
> > 	__builtin_memcpy(dest, src, len);

won't this result in inconsistent copying of src when length of src has changed
by prctl(PR_SET_NAME)?

> > 	return dest;
> > }

This strncpy() assumes that length of src won't change within the function.
I thought prctl(PR_SET_NAME) might break such assumption.

Re: [PATCH 1/1] coredump: use task comm instead of (unknown)

[Andrew Morton]

On Sat, 7 May 2011 11:14:14 +0900 Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> wrote:

> Andrew Morton wrote:
> > > char *strncpy(char *dest, const char *src, size_t n)
> > > {
> > > 	size_t len = __strnend(src, n) - src;
> 
> If src was overwritten by prctl(PR_SET_NAME) at this moment (i.e. after len was
> calculated),
> 
> > > 	__builtin_memset(dest + len, 0, n - len);
> > > 	__builtin_memcpy(dest, src, len);
> 
> won't this result in inconsistent copying of src when length of src has changed
> by prctl(PR_SET_NAME)?
> 
> > > 	return dest;
> > > }
> 
> This strncpy() assumes that length of src won't change within the function.
> I thought prctl(PR_SET_NAME) might break such assumption.

PR_SET_NAME uses set_task_comm() which has appropriate locking to
protect against get_task_comm().  

If kernel code directly accesses task->comm without taking task_lock()
then yes, it's racy.

Re: [PATCH 1/1] coredump: use task comm instead of (unknown)

[Tetsuo Handa]

Andrew Morton wrote:
> PR_SET_NAME uses set_task_comm() which has appropriate locking to
> protect against get_task_comm().  
> 
> If kernel code directly accesses task->comm without taking task_lock()
> then yes, it's racy.

I see. Well,

  grep -r -- '->comm[,;)]' .

in 2.6.39-rc6 shows so many users directly reading/updating, including

	case 'e':
		err = cn_printf(cn, "%s", current->comm);
		break;

in format_corename().

> Hopefully John Stultz will soon be presenting us with a %p modifier for
> displaying task_struct.comm.

That will be nice. But caller might by error call printk() with that modifier
between task_lock() and task_unlock().