Official protobuf resources may distribute unpatched and vulnerable code, and direct users to vulnerable projects
42 views
Skip to first unread message
Fred Weitendorf
Apr 2, 2026, 1:36:47 PM
Apr 2
to Protocol Buffers
I'm posting here because I found a very outdated tool linked in the
protobuf repo that seemed to possibly include unpatched CVEs. Analysis/a
quick fix (just remove the link to the old tool) can be found in https://github.com/protocolbuffers/protobuf/pull/26689
Realizing that the old source code introduced possible vulnerabilities inspired me to dig into that particular CVE: https://www.cve.org/CVERecord?id=CVE-2015-5237 , especially because the 2GB message limit is actually something I've run into in my regular work. In https://github.com/protocolbuffers/protobuf/releases/ I noticed that many of the old official 1P releases may contain unpatched code allowing for buffer overflow/RCE in older C++ proto deserialization implementations that don't enforce the ~2GB max message limit, like https://github.com/protocolbuffers/protobuf/releases/tag/v3.0.2 (see https://nvd.nist.gov/vuln/search#/nvd/home?cpeFilterMode=cpe&cpeName=cpe:2.3:a:google:protobuf:3.0.2:*:*:*:*:*:*:*&resultType=records ). Then I found that projects seem to still be using these vulnerable releases still, like in https://github.com/Wikidata/primarysources/blob/07bf74e56ada68a211f1712cc2473a98ce92ef0c/.travis.yml#L31 .
I'm not sure what the policy is for distributing old packages with vulnerabilities (maybe they're intentionally left up without patches?) but I think there might be a need for some cleanup in https://github.com/protocolbuffers/protobuf to remove broken, outdated, and vulnerable releases and any references to tools/projects that also contain unpatched code.
Right now the protobuf project is linking to 3P tools and unpatched 1P releases with exploitable buffer overflow, as far as I can tell. At the time this was dismissed as not a severe problem because it would only affect messages > 2GB, but I think that was too hasty. Adversarial input could trigger the 2GB overflow by providing large junk string/blob values or tons of repeated entries (possibly from several layers removed upstream, eg like log4j) in any message with repeated or variable-length fields, I think? That would be quite easy to do on any exposed port running a vulnerable grpc service, and probably pretty straightforward to do anywhere else where media or string input could propagate from an untrusted user to an internal grpc service, so I I'm not sure just fixing-forward and leaving the unpatched code up on github for other projects to continue to use was the right call. But I could be mistaken here about something (I hope I am!) so please let me know if so!
Realizing that the old source code introduced possible vulnerabilities inspired me to dig into that particular CVE: https://www.cve.org/CVERecord?id=CVE-2015-5237 , especially because the 2GB message limit is actually something I've run into in my regular work. In https://github.com/protocolbuffers/protobuf/releases/ I noticed that many of the old official 1P releases may contain unpatched code allowing for buffer overflow/RCE in older C++ proto deserialization implementations that don't enforce the ~2GB max message limit, like https://github.com/protocolbuffers/protobuf/releases/tag/v3.0.2 (see https://nvd.nist.gov/vuln/search#/nvd/home?cpeFilterMode=cpe&cpeName=cpe:2.3:a:google:protobuf:3.0.2:*:*:*:*:*:*:*&resultType=records ). Then I found that projects seem to still be using these vulnerable releases still, like in https://github.com/Wikidata/primarysources/blob/07bf74e56ada68a211f1712cc2473a98ce92ef0c/.travis.yml#L31 .
I'm not sure what the policy is for distributing old packages with vulnerabilities (maybe they're intentionally left up without patches?) but I think there might be a need for some cleanup in https://github.com/protocolbuffers/protobuf to remove broken, outdated, and vulnerable releases and any references to tools/projects that also contain unpatched code.
Right now the protobuf project is linking to 3P tools and unpatched 1P releases with exploitable buffer overflow, as far as I can tell. At the time this was dismissed as not a severe problem because it would only affect messages > 2GB, but I think that was too hasty. Adversarial input could trigger the 2GB overflow by providing large junk string/blob values or tons of repeated entries (possibly from several layers removed upstream, eg like log4j) in any message with repeated or variable-length fields, I think? That would be quite easy to do on any exposed port running a vulnerable grpc service, and probably pretty straightforward to do anywhere else where media or string input could propagate from an untrusted user to an internal grpc service, so I I'm not sure just fixing-forward and leaving the unpatched code up on github for other projects to continue to use was the right call. But I could be mistaken here about something (I hope I am!) so please let me know if so!
Em Rauch
Apr 2, 2026, 1:56:19 PM
Apr 2
to Fred Weitendorf, Protocol Buffers
If you think you have found any live security concerns, please report them https://bughunters.google.com/report
I think there's a lot of nuance here on some of the topics that you're raising 🙂
> I'm not sure what the policy is for distributing old packages with vulnerabilities
In general I think no one yanks arbirtarily old end-of-lifed releases even if they are covered by CVEs; a very large number of users use software on trusted inputs and are perfectly happy to continue using end-of-lifed software rather than updating to latest versions and risk compatibility or correctness problems.
Protobuf's version support policy is here
including which major versions are currently supported in each programming language. We do not yank old major versions that are end-of-lifed but also don't patch them (we may patch them in certain exceptional cases).
It is a case that some upstream distros vendor and distribute what amounts to soft-forks of Protobuf since they release and patch as they see fit; most of them stay relatively up to date but it is the case that Ubuntu is distributing C++Proto 3.21 which was end-of-lifed for support from Google in 2024-Q1. But even if they were on latest, Google couldn't necessarily strongly vouch for the correctness of a distro vendored Protobuf release, as they are free to patch as they see fit and its not really within Google's role in the ecosystem to review the patched packages that distros distribute.
> internal grpc service
Worth mentioning here that gRPC enforces a 4 MB limit on messages by default; we generally recommend only raising that incrementally as needed as a security defence-in-depth best practice.
--
You received this message because you are subscribed to the Google Groups "Protocol Buffers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to protobuf+u...@googlegroups.com .
To view this discussion visit https://groups.google.com/d/msgid/protobuf/74ef98b8-fec5-4e39-8d97-c3637070bc73n%40googlegroups.com .
Reply all
Reply to author
Forward
Search
Clear search
Close search
Google apps
Main menu
