KEYS: call_sbin_request_key() must write lock keyrings before modifying them
authorDavid Howells <dhowells@redhat.com>
Fri, 30 Apr 2010 13:32:23 +0000 (14:32 +0100)
committerJames Morris <jmorris@namei.org>
Wed, 5 May 2010 13:50:24 +0000 (23:50 +1000)
commit896903c2f5f79f029388f033a00c3b813bc91201
treef679108ab3c9cda3f5e1f6240afccc6ee3984406
parentf0641cba7729e5e14f82d2eedc398103f5fa31b1
KEYS: call_sbin_request_key() must write lock keyrings before modifying them

call_sbin_request_key() creates a keyring and then attempts to insert a link to
the authorisation key into that keyring, but does so without holding a write
lock on the keyring semaphore.

It will normally get away with this because it hasn't told anyone that the
keyring exists yet.  The new keyring, however, has had its serial number
published, which means it can be accessed directly by that handle.

This was found by a previous patch that adds RCU lockdep checks to the code
that reads the keyring payload pointer, which includes a check that the keyring
semaphore is actually locked.

Without this patch, the following command:

keyctl request2 user b a @s

will provoke the following lockdep warning is displayed in dmesg:

===================================================
[ INFO: suspicious rcu_dereference_check() usage. ]
---------------------------------------------------
security/keys/keyring.c:727 invoked rcu_dereference_check() without protection!

other info that might help us debug this:

rcu_scheduler_active = 1, debug_locks = 0
2 locks held by keyctl/2076:
 #0:  (key_types_sem){.+.+.+}, at: [<ffffffff811a5b29>] key_type_lookup+0x1c/0x71
 #1:  (keyring_serialise_link_sem){+.+.+.}, at: [<ffffffff811a6d1e>] __key_link+0x4d/0x3c5

stack backtrace:
Pid: 2076, comm: keyctl Not tainted 2.6.34-rc6-cachefs #54
Call Trace:
 [<ffffffff81051fdc>] lockdep_rcu_dereference+0xaa/0xb2
 [<ffffffff811a6d1e>] ? __key_link+0x4d/0x3c5
 [<ffffffff811a6e6f>] __key_link+0x19e/0x3c5
 [<ffffffff811a5952>] ? __key_instantiate_and_link+0xb1/0xdc
 [<ffffffff811a59bf>] ? key_instantiate_and_link+0x42/0x5f
 [<ffffffff811aa0dc>] call_sbin_request_key+0xe7/0x33b
 [<ffffffff8139376a>] ? mutex_unlock+0x9/0xb
 [<ffffffff811a5952>] ? __key_instantiate_and_link+0xb1/0xdc
 [<ffffffff811a59bf>] ? key_instantiate_and_link+0x42/0x5f
 [<ffffffff811aa6fa>] ? request_key_auth_new+0x1c2/0x23c
 [<ffffffff810aaf15>] ? cache_alloc_debugcheck_after+0x108/0x173
 [<ffffffff811a9d00>] ? request_key_and_link+0x146/0x300
 [<ffffffff810ac568>] ? kmem_cache_alloc+0xe1/0x118
 [<ffffffff811a9e45>] request_key_and_link+0x28b/0x300
 [<ffffffff811a89ac>] sys_request_key+0xf7/0x14a
 [<ffffffff81052c0b>] ? trace_hardirqs_on_caller+0x10c/0x130
 [<ffffffff81394fb9>] ? trace_hardirqs_on_thunk+0x3a/0x3f
 [<ffffffff81001eeb>] system_call_fastpath+0x16/0x1b

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>
security/keys/request_key.c