Module Observations
Hi,
I was going thr' module code and made some observations:
1. sys_init_module drops the module_mutex semaphore before calling mod->init() function and later reacquires it. After reacquiring, it marks the module state as MODULE_STATE_LIVE.
In the window when mod->init() function is running, isn't it possible that sys_delete_module (running on some other CPU and trying to remove the _same_ module) acquires the module_mutex sem and marks the module state as MODULE_STATE_GOING?
Shouldn't sys_init_module check for
that possibility when it reacquires the semaphore after calling mod->init function?
module.c Fri Jan 2 18:38:57 2004 renamed from module.c.org Fri Jan 2 18:37:54 2004 +2 -1
@@ -1750,7 +1750,8 @@ /* Now it's a first class citizen! */ down(&module_mutex); - mod->state = MODULE_STATE_LIVE; + if (likely(mod->state != MODULE_STATE_GOING)) + mod->state = MODULE_STATE_LIVE; /* Drop initial reference. */ module_put(mod); module_free(mod, mod->module_init);
This off-course means that you are trying to insmod and rmmod the same module simultaneously from different CPUs and hence may not be practical.
2. try_module_get() and module_put()
try_module_get increments the local cpu's ref count for the module
and module_put decrements it.
Is it required that the caller call both these functions from the same CPU? Otherwise, the total refcount for the module will be non-zero!
Signature · 5 lines
Thanks and Regards, Srivatsa Vaddagiri, Linux Technology Center, IBM Software Labs, Bangalore, INDIA - 560017
Quoted 14 lines · On Fri, 2 Jan 2004 18:55:09 +0530 Srivatsa Vaddagiri <[email protected]> wrote:
Hi,
I was going thr' module code and made some observations:1.
sys_init_moduledrops themodule_mutexsemaphore before callingmod->init()function and later reacquires it. After reacquiring, it marks the module state asMODULE_STATE_LIVE.In the window when
mod->init()function is running, isn't it possible thatsys_delete_module(running on some other CPU and trying to remove the _same_ module) acquires themodule_mutexsem and marks the module state asMODULE_STATE_GOING?
Quoted 3 lines
Shouldn't
sys_init_modulecheck for
that possibility when it reacquires the semaphore after callingmod->initfunction?
Good catch. The module removal should refuse to remove it without --force.
I opened this hole when I dropped the sem around mod->init() (because some modules load other modules in their init routine).
Andrew, please apply patch below.
Quoted 7 lines
2.
try_module_get()andmodule_put()
try_module_getincrements the local cpu's ref count for the module
andmodule_putdecrements it.Is it required that the caller call both these functions from the same CPU? Otherwise, the total refcount for the module will be non-zero!
No, it's OK. We only care about the *total* being zero.
Thanks,
Rusty.
Signature · 26 lines
there are those who do and those who hang on and you don't see too
many doers quoting their contemporaries. -- Larry McVoy
Name: Prevent Removal of Module During Init
Author: Rusty Russell
Status: Trivial
D: Vatsa spotted this: you can remove a module while it's being
D: initialized, and that will be bad. Hole was opened when I dropped
D: the sem around the init routine (which can probe for other
D: modules).
diff -urpN --exclude TAGS -X /home/rusty/devel/kernel/kernel-patches/current-dontdiff --minimal .22325-linux-2.6.1-rc1/kernel/module.c .22325-linux-2.6.1-rc1.updated/kernel/module.c
--- .22325-linux-2.6.1-rc1/kernel/module.c 2003-11-24 15:42:33.000000000 +1100
+++ .22325-linux-2.6.1-rc1.updated/kernel/module.c 2004-01-03 15:59:54.000000000 +1100
@@ -687,8 +687,8 @@ sys_delete_module(const char __user *nam
goto out;
}
- /* Already dying? */
- if (mod->state == MODULE_STATE_GOING) {
+ /* Doing init or already dying? */
+ if (mod->state != MODULE_STATE_LIVE) {
/* FIXME: if (force), slam module count and wake up
waiter --RR */
DEBUGP("%s already dying\n", mod->name);Quoted 14 lines · On Fri, 2 Jan 2004 18:55:09 +0530 Srivatsa Vaddagiri <[email protected]> wrote:
Hi,
I was going thr' module code and made some observations:1.
sys_init_moduledrops themodule_mutexsemaphore before callingmod->init()function and later reacquires it. After reacquiring, it marks the module state asMODULE_STATE_LIVE.In the window when
mod->init()function is running, isn't it possible thatsys_delete_module(running on some other CPU and trying to remove the _same_ module) acquires themodule_mutexsem and marks the module state asMODULE_STATE_GOING?
Quoted 3 lines
Shouldn't
sys_init_modulecheck for
that possibility when it reacquires the semaphore after callingmod->initfunction?
Good catch. The module removal should refuse to remove it without --force.
I opened this hole when I dropped the sem around mod->init() (because some modules load other modules in their init routine).
Andrew, please apply patch below.
Quoted 7 lines
2.
try_module_get()andmodule_put()
try_module_getincrements the local cpu's ref count for the module
andmodule_putdecrements it.Is it required that the caller call both these functions from the same CPU? Otherwise, the total refcount for the module will be non-zero!
No, it's OK. We only care about the *total* being zero.
Thanks,
Rusty.
Signature · 26 lines
there are those who do and those who hang on and you don't see too
many doers quoting their contemporaries. -- Larry McVoy
Name: Prevent Removal of Module During Init
Author: Rusty Russell
Status: Trivial
D: Vatsa spotted this: you can remove a module while it's being
D: initialized, and that will be bad. Hole was opened when I dropped
D: the sem around the init routine (which can probe for other
D: modules).
diff -urpN --exclude TAGS -X /home/rusty/devel/kernel/kernel-patches/current-dontdiff --minimal .22325-linux-2.6.1-rc1/kernel/module.c .22325-linux-2.6.1-rc1.updated/kernel/module.c
--- .22325-linux-2.6.1-rc1/kernel/module.c 2003-11-24 15:42:33.000000000 +1100
+++ .22325-linux-2.6.1-rc1.updated/kernel/module.c 2004-01-03 15:59:54.000000000 +1100
@@ -687,8 +687,8 @@ sys_delete_module(const char __user *nam
goto out;
}
- /* Already dying? */
- if (mod->state == MODULE_STATE_GOING) {
+ /* Doing init or already dying? */
+ if (mod->state != MODULE_STATE_LIVE) {
/* FIXME: if (force), slam module count and wake up
waiter --RR */
DEBUGP("%s already dying\n", mod->name);