[PATCH RFC] try_module_get usage

[PATCH RFC] try_module_get usage

[Fernando Luis Vázquez Cao]

I keep seeing uses of try_module_get(THIS_MODULE) which seem to mimic
the behavior of the former MOD_INC_USE_COUNT. The UBI driver is one
example:

int ubi_get_device_info(int ubi_num, struct ubi_device_info *di)
{
        const struct ubi_device *ubi;

        if (!try_module_get(THIS_MODULE))
                return -ENODEV;

        if (ubi_num < 0 || ubi_num >= UBI_MAX_DEVICES ||
            !ubi_devices[ubi_num]) {
                module_put(THIS_MODULE);
                return -ENODEV;
        }

        ubi = ubi_devices[ubi_num];
        di->ubi_num = ubi->ubi_num;
        di->leb_size = ubi->leb_size;
        di->min_io_size = ubi->min_io_size;
        di->ro_mode = ubi->ro_mode;
        di->cdev = MKDEV(ubi->major, 0);
        module_put(THIS_MODULE);
        return 0;
}

My understanding is that this is not completely safe (we could be
preempted before try_modules_get gets executed) and that it is the
caller who should manipulate the refcounts. Am I missing something here?

Thank you in advance.

[PATCH] UBI: cleanup usage of try_module_get

[Fernando Luis Vázquez Cao]

The use of try_module_get(THIS_MODULE) in ubi_get_device_info does not
offer real protection against unexpected driver unloads, since we could
be preempted before try_modules_get gets executed. It is the caller who
should manipulate the refcounts. Besides, ubi_get_device_info is an
exported symbol which guarantees protection when accessed through
symbol_get.

Signed-off-by: Fernando Luis Vazquez Cao <fernando@oss.ntt.co.jp>
---

diff -urNp linux-2.6.22-orig/drivers/mtd/ubi/kapi.c linux-2.6.22/drivers/mtd/ubi/kapi.c
--- linux-2.6.22-orig/drivers/mtd/ubi/kapi.c	2007-07-09 08:32:17.000000000 +0900
+++ linux-2.6.22/drivers/mtd/ubi/kapi.c	2007-07-11 16:34:10.000000000 +0900
@@ -37,12 +37,8 @@ int ubi_get_device_info(int ubi_num, str
 {
 	const struct ubi_device *ubi;
 
-	if (!try_module_get(THIS_MODULE))
-		return -ENODEV;
-
 	if (ubi_num < 0 || ubi_num >= UBI_MAX_DEVICES ||
 	    !ubi_devices[ubi_num]) {
-		module_put(THIS_MODULE);
 		return -ENODEV;
 	}
 
@@ -52,7 +48,6 @@ int ubi_get_device_info(int ubi_num, str
 	di->min_io_size = ubi->min_io_size;
 	di->ro_mode = ubi->ro_mode;
 	di->cdev = MKDEV(ubi->major, 0);
-	module_put(THIS_MODULE);
 	return 0;
 }
 EXPORT_SYMBOL_GPL(ubi_get_device_info);

Re: [PATCH] UBI: cleanup usage of try_module_get

[Artem Bityutskiy]

On Wed, 2007-07-11 at 17:11 +0900, Fernando Luis Vázquez Cao wrote:
> The use of try_module_get(THIS_MODULE) in ubi_get_device_info does not
> offer real protection against unexpected driver unloads, since we could
> be preempted before try_modules_get gets executed. It is the caller who
> should manipulate the refcounts. Besides, ubi_get_device_info is an
> exported symbol which guarantees protection when accessed through
> symbol_get.

Yeah, looks fair, committed to UBI git, thanks.

-- 
Best regards,
Artem Bityutskiy (Битюцкий Артём)

Re: [PATCH RFC] try_module_get usage

[Christoph Hellwig]

On Wed, Jul 11, 2007 at 04:55:31PM +0900, Fernando Luis V?zquez Cao wrote:
> I keep seeing uses of try_module_get(THIS_MODULE) which seem to mimic
> the behavior of the former MOD_INC_USE_COUNT. The UBI driver is one
> example:

It's of course buggy as hell.  But UBI folks prefer not to listen to
advice and keep their bugs.

Re: [PATCH RFC] try_module_get usage

[Artem Bityutskiy]

On Wed, 2007-07-11 at 10:41 +0100, Christoph Hellwig wrote:
> On Wed, Jul 11, 2007 at 04:55:31PM +0900, Fernando Luis V?zquez Cao
> wrote:
> > I keep seeing uses of try_module_get(THIS_MODULE) which seem to
> mimic
> > the behavior of the former MOD_INC_USE_COUNT. The UBI driver is one
> > example:
> 
> It's of course buggy as hell.  But UBI folks prefer not to listen to
> advice and keep their bugs.

Christoph, stop this please, point the bugs we do not listen about. Your
patches sit in the UBI git and wait for the merge window:

http://git.infradead.org/?p=ubi-2.6.git;a=summary

I plan to let dwmw2 pull them and send to Linus together with other
MTD/JFFS2 patches. If you have specific requests - go ahead.

-- 
Best regards,
Artem Bityutskiy (Битюцкий Артём)

Re: [PATCH RFC] try_module_get usage

[Rusty Russell]

On Wed, 2007-07-11 at 16:55 +0900, Fernando Luis Vázquez Cao wrote:
> I keep seeing uses of try_module_get(THIS_MODULE) which seem to mimic
> the behavior of the former MOD_INC_USE_COUNT. The UBI driver is one
> example:
> 
> int ubi_get_device_info(int ubi_num, struct ubi_device_info *di)
> {
>         const struct ubi_device *ubi;
> 
>         if (!try_module_get(THIS_MODULE))
>                 return -ENODEV;

Yes, this code is wrong.

Thanks,
Rusty.