On Mon, 13 Feb 2023 16:57:17 GMT, Severin Gehwolf <sgehw...@openjdk.org> wrote:
> The `jimage` location attributes are terminated with `ATTRIBUTE_END`-kinds. > However, > the byte containing `ATTRIBUTE_END` (most significant 5 bits, represent > `kind`), might > be non-zero in the lower 3 bits (values up to `0x07` represent > `ATTRIBUTE_END`). The JDK code > handles this case correctly in > [`ImageLocation.decompress()`](https://github.com/openjdk/jdk/blob/master/src/java.base/share/classes/jdk/internal/jimage/ImageLocation.java#L69..L71). > However, the `libjimage` > code in `java.base` doesn't. That can lead to segfaults reading random bytes > and offsets. > > I propose to break the loop if `ATTRIBUTE_END` is being encountered so that > reading stops. > > Testing: > - [x] `test/jdk/tools/jimage` and `test/jdk/jdk/internal/jimage` tests. > - [x] Manual testing with a patched JDK to write non-zero bytes containing > `ATTRIBUTE_END` into the jimage. Segfaults before, runs fine after. > - [x] GHA. > > Thoughts? This pull request has now been integrated. Changeset: ee5f6e15 Author: Severin Gehwolf <sgehw...@openjdk.org> URL: https://git.openjdk.org/jdk/commit/ee5f6e156de0fd3d78adf60951866f43c492725b Stats: 3 lines in 1 file changed: 3 ins; 0 del; 0 mod 8302337: JDK crashes if lib/modules contains non-zero byte containing ATTRIBUTE_END Reviewed-by: stuefe, jlaskey, alanb ------------- PR: https://git.openjdk.org/jdk/pull/12539