On Thu, Feb 17, 2022 at 10:30:13AM -0600, Eric Blake wrote:
On Thu, Feb 17, 2022 at 02:36:43PM +0000, Richard W.M. Jones wrote:
This filter doesn't call the next_open function in the non-TLS case, and therefore it never opens the plugin. This leaves the internal state of nbdkit a bit strange. There is no plugin context allocated, and the last filter in the chain has a context c_next pointer of NULL.
This works, provided we intercept every possible callback, check the non-TLS case, and prevent it from calling the next function (because it would dereference the NULL c_next).
To avoid a crash in backend_block_size we must therefore provide a .block_size callback in this filter. --- filters/tls-fallback/tls-fallback.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+)
ACK.
Would you like to squash this in, and/or have me commit this separately?
I was actually thinking about squashing my patches 1-4 together. They're all really the same change, but I kept them separate for ease of review. What do you think? But I think this patch:
commit 8c00ca2fe418aeecf0818feed227a72e76d87f18 Author: Eric Blake <eblake@redhat.com> Date: Thu Feb 17 10:24:50 2022 -0600
tls-fallback: Enhance comments about required callbacks
Rich ran into a crash while trying to add a new .block_size callback reachable during the client connection handshake (via .prepare), and nearly worked around the crash by reintroducing CVE-2019-14850 instead of the proper fix of implementing the new callback. Add some comments to help us avoid a similar regression the next time we add a callback.
diff --git a/filters/tls-fallback/tls-fallback.c b/filters/tls-fallback/tls-fallback.c index fab9e58b..64e84926 100644 --- a/filters/tls-fallback/tls-fallback.c +++ b/filters/tls-fallback/tls-fallback.c @@ -107,7 +107,11 @@ tls_fallback_open (nbdkit_next_open *next, nbdkit_context *nxdata, const char *exportname, int is_tls) { /* We do NOT want to call next() when insecure, because we don't - * know how long it will take. + * know how long it will take. See also CVE-2019-14850 in + * nbdkit-security.pod. But that means that this filter must + * override every possible callback that can be reached during + * handshake, to avoid passing through a non-TLS call to a missing + * backend. */ if (!is_tls) return &message; /* See NOT_TLS for this choice of handle */
... would stay separate, and you can push it before or after. Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com Fedora Windows cross-compiler. Compile Windows programs, test, and build Windows installers. Over 100 libraries supported. http://fedoraproject.org/wiki/MinGW