NFS/sunrpc: don't use a credential with extra groups.
Subsystems (3)
Hi Neil
I queued this patch up for linux-next testing earlier this morning. I could have included it in the pull I sent to Linus, but I'm trying to be a bit more strict about integration testing prior to pushing.
I also added a Cc: [email protected]. I'm assuming that was appropriate.
Cheers
Trond
NeilBrown <[email protected]> wrote:
Hi Trond,
A customer hit this bug recently - which seems to have been around forever - at least since 2.6.16.
I suspect few people would risk hitting it, but our customer does interesting things with group lists (presumably some least-privilege policy) and they managed to trigger it for us.
NeilBrown
Quoted 1 line
From
80b7771c0e18f755a6b9679e6c3b9d1449ba40eeMon Sep 17 00:00:00 2001
From: NeilBrown <[email protected]>
Date: Tue, 25 Oct 2011 10:17:06 +1100
Subject: [PATCH] NFS/sunrpc: don't use a credential with extra groups.
The sunrpc layer keeps a cache of recently used credentials and 'unx_match' is used to find the credential which matches the current process.
However unx_match allows a match when the cached credential has extra groups at the end of uc_gids list which are not in the process group list.
So if a process with a list of (say) 4 group accesses a file and gains access because of the last group in the list, then another process with the same uid and gid, and a gid list being the first tree of the gids of the original process tries to access the file, it will be granted access even though it shouldn't as the wrong rpc credential will be used.
--- net/sunrpc/auth_unix.c | 3 +++ 1 files changed, 3 insertions(+), 0 deletions(-)
net/sunrpc/auth_unix.c +3 -0
@@ -129,6 +129,9 @@ unx_match(struct auth_cred *acred, struct rpc_cred *rcred, int flags)for (i =0 ; i < groups ; i++)if (cred->uc_gids[i] !=GROUP_AT (acred->group_info, i))return 0 ; +if (groups <NFS_NGROUPS && + cred->uc_gids[groups] !=NOGROUP ) +return 0 ;return 1 ; }
Signature · 1 line
1.7.7
Quoted 5 lines · Trond
Hi Neil
I queued this patch up for linux-next testing earlier this morning. I could have included it in the pull I sent to Linus, but I'm trying to be a bit more strict about integration testing prior to pushing.
I also added a Cc: [email protected]. I'm assuming that was appropriate.
Thanks.
I was in two minds about -stable as it requires special care to trigger the bug. I am certainly not against it though.
(and I fully support your more strict approach to integration testing! I'll take it as an example to aspire to).
Thanks,
NeilBrown