Skip to contentSkip to conversation

module: set MODULE_STATE_GOING state when a module fails to load

PATCH Applied Miroslav Benes 2 participants 4 messages Oct 27, 2020 – Oct 29, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 module: set MODULE_STATE_GOING state when a module fails to load

If a module fails to load due to an error in prepare_coming_module(), the following error handling in load_module() runs with MODULE_STATE_COMING in module's state. Fix it by correctly setting MODULE_STATE_GOING under "bug_cleanup" label.

---
 kernel/module.c | 1 +
 1 file changed, 1 insertion(+)
kernel/module.c +1 -0
index a4fa44a652a7..b34235082394 100644
@@ -3991,6 +3991,7 @@ static int load_module(struct load_info *info, const char __user *uargs,
 				     MODULE_STATE_GOING, mod);
 	klp_module_going(mod);
  bug_cleanup:
+	mod->state = MODULE_STATE_GOING;
 	/* module_bug_cleanup needs module_mutex protection */
 	mutex_lock(&module_mutex);
 	module_bug_cleanup(mod);
Signature · 1 line
2.29.0
#1 lore raw

+++ Miroslav Benes [27/10/20 15:03 +0100]:

Quoted 15 lines

If a module fails to load due to an error in prepare_coming_module(), the following error handling in load_module() runs with MODULE_STATE_COMING in module's state. Fix it by correctly setting MODULE_STATE_GOING under "bug_cleanup" label.

---
kernel/module.c | 1 +

1 file changed, 1 insertion(+)
kernel/module.c +0 -0
index a4fa44a652a7..b34235082394 100644

kernel/module.c static int load_module(struct load_info *info, const char __user *uargs, · line 3997

Quoted 7 lines
kernel/module.c +1 -0
@@ -3991,6 +3991,7 @@ static int load_module(struct load_info *info, const char __user *uargs,
 				     MODULE_STATE_GOING, mod);
 	klp_module_going(mod);
 bug_cleanup:
+	mod->state = MODULE_STATE_GOING;
 	/* module_bug_cleanup needs module_mutex protection */
 	mutex_lock(&module_mutex);
 	module_bug_cleanup(mod);

Thanks for the fix! Hmm, I am wondering if we also need to set the module to GOING if it happens to fail while it is still UNFORMED.

Currently, when a module is UNFORMED and encounters an error during load_module(), it stays UNFORMED until it finally goes away. That sounds fine, but try_module_get() technically permits you to get a module while it's UNFORMED (but not if it's GOING). Theoretically someone could increase the refcount of an unformed module that has encountered an error condition and is in the process of going away. This shouldn't happen if we properly set the module to GOING whenever it encounters an error during load_module().

But - I cannot think of a scenario where someone could call try_module_get() on an unformed module, since find_module() etc. do not return unformed modules, so they shouldn't be visible outside of the module loader. So in practice, I think we're probably safe here..

#2 lore raw
Quoted 28 lines · Jessica Yu

+++ Miroslav Benes [27/10/20 15:03 +0100]:

Quoted 22 lines

If a module fails to load due to an error in prepare_coming_module(), the following error handling in load_module() runs with MODULE_STATE_COMING in module's state. Fix it by correctly setting MODULE_STATE_GOING under "bug_cleanup" label.

---
kernel/module.c | 1 +

1 file changed, 1 insertion(+)
kernel/module.c +1 -0
index a4fa44a652a7..b34235082394 100644
@@ -3991,6 +3991,7 @@ static int load_module(struct load_info *info, const
 char __user *uargs,
 			     MODULE_STATE_GOING, mod);
 	klp_module_going(mod);
 bug_cleanup:
+	mod->state = MODULE_STATE_GOING;
 /* module_bug_cleanup needs module_mutex protection */
 mutex_lock(&module_mutex);
Quoted 1 line

module_bug_cleanup(mod);

Thanks for the fix! Hmm, I am wondering if we also need to set the module to GOING if it happens to fail while it is still UNFORMED.

Quoted 6 lines

Currently, when a module is UNFORMED and encounters an error during load_module(), it stays UNFORMED until it finally goes away. That sounds fine, but try_module_get() technically permits you to get a module while it's UNFORMED (but not if it's GOING). Theoretically someone could increase the refcount of an unformed module that has encountered an error condition and is in the process of going away.

Right.

Quoted 2 lines

This shouldn't happen if we properly set the module to GOING whenever it encounters an error during load_module().

That's correct.

Quoted 4 lines

But - I cannot think of a scenario where someone could call try_module_get() on an unformed module, since find_module() etc. do not return unformed modules, so they shouldn't be visible outside of the module loader. So in practice, I think we're probably safe here..

Hopefully yes. I haven't found anything that would contradict it.

I think it is even safer to leave UNFORMED there. free_module() explicitly sets UNFORMED state too while going through the similar process.

ftrace_release_mod() is the only inconsistency there. It is called with UNFORMED in load_module() if going through ddebug_cleanup label directly, and with GOING in both do_init_module() before free_module() is called and delete_module syscall. But it probably does not care.

Miroslav

Applied
#3 lore raw

+++ Miroslav Benes [29/10/20 13:31 +0100]:

Quoted 49 lines · Jessica Yu
Quoted 29 lines

+++ Miroslav Benes [27/10/20 15:03 +0100]:

Quoted 22 lines

If a module fails to load due to an error in prepare_coming_module(), the following error handling in load_module() runs with MODULE_STATE_COMING in module's state. Fix it by correctly setting MODULE_STATE_GOING under "bug_cleanup" label.

---
kernel/module.c | 1 +

1 file changed, 1 insertion(+)
kernel/module.c +1 -0
index a4fa44a652a7..b34235082394 100644
@@ -3991,6 +3991,7 @@ static int load_module(struct load_info *info, const
 char __user *uargs,
 			     MODULE_STATE_GOING, mod);
 	klp_module_going(mod);
 bug_cleanup:
+	mod->state = MODULE_STATE_GOING;
 /* module_bug_cleanup needs module_mutex protection */
 mutex_lock(&module_mutex);
Quoted 1 line

module_bug_cleanup(mod);

Thanks for the fix! Hmm, I am wondering if we also need to set the module to GOING if it happens to fail while it is still UNFORMED.

Quoted 6 lines

Currently, when a module is UNFORMED and encounters an error during load_module(), it stays UNFORMED until it finally goes away. That sounds fine, but try_module_get() technically permits you to get a module while it's UNFORMED (but not if it's GOING). Theoretically someone could increase the refcount of an unformed module that has encountered an error condition and is in the process of going away.

Right.

Quoted 2 lines

This shouldn't happen if we properly set the module to GOING whenever it encounters an error during load_module().

That's correct.

Quoted 4 lines

But - I cannot think of a scenario where someone could call try_module_get() on an unformed module, since find_module() etc. do not return unformed modules, so they shouldn't be visible outside of the module loader. So in practice, I think we're probably safe here..

Quoted 4 lines

Hopefully yes. I haven't found anything that would contradict it.

I think it is even safer to leave UNFORMED there. free_module() explicitly sets UNFORMED state too while going through the similar process.

That's true, and agreed.

Quoted 4 lines

ftrace_release_mod() is the only inconsistency there. It is called with UNFORMED in load_module() if going through ddebug_cleanup label directly, and with GOING in both do_init_module() before free_module() is called and delete_module syscall. But it probably does not care.

Yes, I guess that inconsistency with ftrace_release_mod() has been there long before this patch :-/ A quick look through ftrace.c doesn't suggest to me there are any direct dependencies on module state there (other than that comment above ftrace_module_init()). In practice there hasn't been any issues..

I have applied this to modules-next. Thanks!

Jessica