[crac] RFR: Ignore open files in /var/lib/sss/mc

Roman Marchenko rmarchenko at openjdk.org
Thu Nov 2 16:07:45 UTC 2023


On Thu, 2 Nov 2023 14:58:13 GMT, Radim Vansa <rvansa at openjdk.org> wrote:

> I was considering different ways to fix this - there are actually two problems:
> 1) how to detect that?
> * This PR whitelisst all files in the directory (we could explicitly name `passwd`, `group`, `sid` and `initgroups`...)
> * We could manually check all places in JDK that call `getpwuid*`, `getpwname*`, `getgrgid*`and `getgrname*` and maybe some other functions, and diff FDs opened before/after the call. However this a) has performance impact b) is prone to races
> * Intercept the call: either catch syscalls (ptrace or seccomp), or patching `sss_open_cloexec` in memory, or preloading it? Rather complicated.
> 2) what to do with the open FD?
> * Current solution is to leave this up to CRIU (or another C/R engine). Again the simplest
> * We could close this; the FD is `fstat`-validated later on, so this would cause errors from these functions. We would need to patch the library, de-initializing the implementation (also risky).
> 
> Normally I would try to include a testcase but given that this is system-dependent (and the issue does not appear even in CentOS Stream 9 container) I've only did manual testing.

src/hotspot/os/linux/crac_linux.cpp line 336:

> 334:     // or other functions in this family.
> 335: #define SSS_NSS_MEMCACHE "/var/lib/sss/mc/"
> 336:     if (!strncmp(details, SSS_NSS_MEMCACHE, sizeof(SSS_NSS_MEMCACHE) - 1)) {

I believe GCC can manage with `const char *` and `strlen` call in compile time here. `constexpr` could be added as well, but not necessary I guess.

-------------

PR Review Comment: https://git.openjdk.org/crac/pull/137#discussion_r1380389278


More information about the crac-dev mailing list