Re: [PATCH v2] module: add usage links when calling ref_module func

[Jessica Yu <jeyu@kernel.org>]

+++ Zhiqiang Liu [03/07/19 10:09 +0800]:
>From: Zhiqiang Liu <liuzhiqiang26@huawei.com>
>
>Users can call ref_module func in their modules to construct
>relationships with other modules. However, the holders
>'/sys/module/<mod-name>/holders' of the target module donot include
>the users` module. So lsmod command misses detailed info of 'Used by'.
>
>When load module, the process is given as follows,
>load_module()
>	-> mod_sysfs_setup()
>		-> add_usage_links
>	-> do_init_module
>		-> mod->init()
>
>add_usage_links func creates holders of target modules linking to
>this module. If ref_module is called in mod->init() func, the usage
>links cannot be added.
>
>Here, we will add usage link of a to b's holder_dir.
>
>V1->V2:
>- remove incorrect Fixes tag
>- fix error handling of sysfs_create_link as suggested by Jessica Yu
>
>Signed-off-by: Zhiqiang Liu <liuzhiqiang26@huawei.com>
>Suggested-by: Jessica Yu <jeyu@kernel.org>
>Reviewed-by: Kang Zhou <zhoukang7@huawei.com>
>---
> kernel/module.c | 18 ++++++++++++++----
> 1 file changed, 14 insertions(+), 4 deletions(-)
>
>diff --git a/kernel/module.c b/kernel/module.c
>index 80c7c09584cf..672abce2222c 100644
>--- a/kernel/module.c
>+++ b/kernel/module.c
>@@ -837,25 +837,26 @@ static int already_uses(struct module *a, struct module *b)
>  *    'b' can walk the list to see who sourced them), and of 'a'
>  *    targets (so 'a' can see what modules it targets).
>  */
>-static int add_module_usage(struct module *a, struct module *b)
>+static struct module_use *add_module_usage(struct module *a, struct module *b)
> {
> 	struct module_use *use;
>
> 	pr_debug("Allocating new usage for %s.\n", a->name);
> 	use = kmalloc(sizeof(*use), GFP_ATOMIC);
> 	if (!use)
>-		return -ENOMEM;
>+		return NULL;
>
> 	use->source = a;
> 	use->target = b;
> 	list_add(&use->source_list, &b->source_list);
> 	list_add(&use->target_list, &a->target_list);
>-	return 0;
>+	return use;
> }
>
> /* Module a uses b: caller needs module_mutex() */
> int ref_module(struct module *a, struct module *b)
> {
>+	struct module_use *use;
> 	int err;
>
> 	if (b == NULL || already_uses(a, b))
>@@ -866,9 +867,18 @@ int ref_module(struct module *a, struct module *b)
> 	if (err)
> 		return err;
>
>-	err = add_module_usage(a, b);
>+	use = add_module_usage(a, b);
>+	if (!use) {
>+		module_put(b);
>+		return -ENOMEM;
>+	}
>+
>+	err = sysfs_create_link(b->holders_dir, &a->mkobj.kobj, a->name);

Sigh. This ultimately doesn't work because in load_module(), we use
ref_module() in resolve_symbol(), and mod->mkobj.kobj doesn't get
initialized until mod_sysfs_init(), which happens much later in
load_module(). So what happens is that the ref_module(mod, owner) call
in resolve_symbol() returns an error because sysfs_create_link() fails here.
We could *maybe* move sysfs initialization earlier in load_module()
but that is an entirely untested idea and I would need to think about
that more.

> 	if (err) {
> 		module_put(b);
>+		list_del(&use->source_list);
>+		list_del(&use->target_list);
>+		kfree(use);
> 		return err;
> 	}
> 	return 0;
>-- 
>2.19.1
>