Skip to contentSkip to conversation

cuse: wait for pending RCU callbacks on module exit

PATCH Applied Baokun Li 2 participants 4 messages Aug 14 – Aug 18

Subsystems (2)
lore
Share & tools
lore

Baokun Li

author
#0 lore raw

PATCH 1/1 cuse: wait for pending RCU callbacks on module exit

Since commit 053fc4f755ad ("fuse: fix UAF in rcu pathwalks"), fuse_conn_put() frees the fuse_conn through call_rcu() rather than synchronously. For cuse, fc->release is cuse_fc_release(), which lives in the cuse module. If the module is removed before the RCU grace period ends, the callback jumps into freed module memory:

userspace / module unload | RCU softirq
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 close(/dev/cuse)                    |
  cuse_channel_release()             |
   fuse_dev_release()                |
    fuse_conn_put(fch->conn)         |
     call_rcu(delayed_release) ------+---> callback queued
                                     |
 rmmod cuse                          |
  cuse_exit()                        |
   cuse_channel_destroy()            |
   ...                               |
   return                            |
                                     |
 <module text freed>                 |
                                     |  rcu_do_batch()
                                     |   delayed_release()
                                     |    fc->release()
                                     |     -> cuse_fc_release()
                                     |        ^^^ freed text!

The freed module text is unmapped by vfree(), so the jump into the stale callback triggers a page-fault Oops. If the virtual address is subsequently reused, the callback could execute unrelated code (undefined behaviour).

Fix this by calling rcu_barrier() in cuse_exit() so that any pending fuse_conn release callback completes before the module is removed.

Fixes 053fc4f755ad ("fuse: fix UAF in rcu pathwalks")
---
 fs/fuse/cuse.c | 5 +++++
 1 file changed, 5 insertions(+)
fs/fuse/cuse.c +5 -0
index 3c15b5ba16d7..1c71783ca5a5 100644
@@ -653,6 +653,11 @@ static void __exit cuse_exit(void)
 {
 	misc_deregister(&cuse_miscdev);
 	class_destroy(cuse_class);
+	/*
+	 * Wait for pending call_rcu() callbacks that call back into
+	 * this module via fc->release (cuse_fc_release).
+	 */
+	rcu_barrier();
 }
 
 module_init(cuse_init);
Signature · 1 line
2.43.7
maintainer
#1 lore raw
Quoted 6 lines · Baokun Li

Since commit 053fc4f755ad ("fuse: fix UAF in rcu pathwalks"), fuse_conn_put() frees the fuse_conn through call_rcu() rather than synchronously. For cuse, fc->release is cuse_fc_release(), which lives in the cuse module. If the module is removed before the RCU grace period ends, the callback jumps into freed module memory:

free_module() calls synchronize_rcu(), so this should not be an issue.

Thanks,
Miklos

author
#2 lore raw

Hi Miklos,

Quoted 7 lines · Miklos Szeredi
Quoted 5 lines · Baokun Li

Since commit 053fc4f755ad ("fuse: fix UAF in rcu pathwalks"), fuse_conn_put() frees the fuse_conn through call_rcu() rather than synchronously. For cuse, fc->release is cuse_fc_release(), which lives in the cuse module. If the module is removed before the RCU grace period ends, the callback jumps into freed module memory:

free_module() calls synchronize_rcu(), so this should not be an issue.

synchronize_rcu() only waits for pre-existing RCU read-side critical sections to complete. It does not wait for pending call_rcu() callbacks to execute. The callback can still be queued when module text is freed. The rcu_barrier() is needed to flush them.

Many modules (including fuse itself) already call rcu_barrier() on exit for exactly this reason. Documentation/RCU/rcubarrier.rst also explicitly states that synchronize_rcu() does not wait for callbacks and cannot replace rcu_barrier().

Thanks,
Baokun

maintainer Applied
#3 lore raw
Quoted 4 lines · Baokun Li

Many modules (including fuse itself) already call rcu_barrier() on exit for exactly this reason. Documentation/RCU/rcubarrier.rst also explicitly states that synchronize_rcu() does not wait for callbacks and cannot replace rcu_barrier().

Got it.

Applied, thanks.

Miklos